2026-05-10 23:51:24 +02:00
import type { KtxLlmProvider } from '@ktx/llm' ;
import { generateKtxText } from '../llm/index.js' ;
2026-05-10 23:12:26 +02:00
import type {
2026-05-10 23:51:24 +02:00
KtxColumnSampleInput ,
KtxColumnSampleResult ,
KtxScanContext ,
KtxScanLoggerPort ,
fix(context): merge overlay columns onto manifest columns by name (#94)
* fix(context): merge overlay columns onto manifest columns by name
composeOverlay was appending overlay columns to the manifest column list,
producing duplicate entries when dbt/metabase overlays declared a column
just to attach descriptions. The duplicates carried no `type`, so the
pydantic SourceDefinition rejected them at semantic-query time and broke
`ktx sl query` for every overlay-backed measure. Now overlay columns
match base columns by name (case-insensitive): same-name entries merge
onto the manifest (overlay fields win, type/role fall back to the base,
descriptions merge per source key) and only new names append.
* refactor(sl): split overlay columns from column_overrides and enforce TS/Python wire contract
Overlay sources now have two distinct collections: `columns:` for computed
columns (requiring `expr` + `type`) and `column_overrides:` for metadata
patches to inherited manifest columns. Composing or loading an overlay that
mixes the two — or references an unknown column — fails with a typed error.
Introduce `ResolvedSemanticLayerSource` / `resolvedSourceSchema` /
`toResolvedWire` as the strict shape sent to the Python engine, and add a
schema contract test that diffs Zod against the Pydantic JSON schema dumped
by `python -m semantic_layer dump-schema`. `SourceDefinition` is now
`extra="forbid"` on the Python side.
`loadAllSources` surfaces per-file load errors instead of swallowing them,
so validation/query paths can report manifest shard parse failures.
* fix(context): make scan description generation resilient and quiet
A transient sampleTable failure during ingest used to take out every
table in a connection: generateTableDescription returned a hardcoded
'Table not found' string into descriptions.ai, and KtxDescriptionGenerator
was constructed without a logger, so the failure left no trail anywhere.
- sampleTable / sampleColumn calls retry 3x with 200/400/800ms backoff,
honouring KtxScanContext.signal via a new KtxAbortedError.
- On retry exhaustion or missing capability, table generation falls back
to a metadata-only prompt built from column name / native type / comment
/ rawDescriptions. The column path follows the same rule -- call the
LLM when any of samples or rawDescriptions are available; skip only
when both are absent.
- Logger is now threaded from KtxScanContext into the generator. Failures
emit structured KtxScanWarning entries (new description_fallback_used
code, plus existing sampling_failed / enrichment_failed /
connector_capability_missing). ktx scan groups warnings by code so a
batch of identical failures collapses to one summary line plus sample.
- Returns null on failure instead of the 'Table not found' sentinel; the
manifest writer's existing guard already skips empty descriptions, so
schema YAML no longer carries misleading text. SCAN_MANAGED_DESCRIPTION_KEYS
already strips stale 'ai' on merge, so existing YAML clears on next run.
Also suppress AI SDK v6 'system in messages' warning: pull system messages
out of KtxMessageBuilder.wrapSimple's output via a new splitKtxSystemMessages
helper and pass them top-level to generateText (preserves cacheControl
providerOptions on the SystemModelMessage). Agent-runner's local
splitSystemPromptMessages dedupes onto the shared helper.
* test(docs): align examples-docs assertions with revamped docs
PR #103 (setup/guide doc revamp) reworded several CLI examples and
connection labels; the assertions in scripts/examples-docs.test.mjs
still referenced the pre-revamp wording and were failing in CI on main.
Update the regexes to match the post-revamp content:
- drop the `--json` flag from the sl-query example expectation
- move the `Driver:` / `Status: ok` probe to the connection reference,
which is where that output now lives (driver id is lowercase
`postgres`, not the display name `PostgreSQL`)
- drop the obsolete `Install \`uv\`...` troubleshooting line
- accept `<connectionId>` everywhere; the docs no longer use the
hyphenated `<connection-id>` form
- match the `warehouse` connection id used in the quickstart instead of
the `postgres-warehouse` id only used in the README and setup ref
* fix(sl): skip TS/Python schema contract test when uv is unavailable
The TypeScript checks CI job does not install uv or Python, so the
module-level `execFileSync('uv', ...)` in schemas.contract.test.ts threw
ENOENT and failed the suite. Wrap the schema dump in a try/catch and
guard the describe block with `describe.skipIf` so the test skips in
environments without uv. Local dev and any CI job that has uv on PATH
still runs the cross-language contract assertion.
2026-05-15 02:11:04 +02:00
KtxScanWarning ,
2026-05-10 23:51:24 +02:00
KtxTableRef ,
KtxTableSampleInput ,
KtxTableSampleResult ,
2026-05-10 23:12:26 +02:00
} from './types.js' ;
fix(context): merge overlay columns onto manifest columns by name (#94)
* fix(context): merge overlay columns onto manifest columns by name
composeOverlay was appending overlay columns to the manifest column list,
producing duplicate entries when dbt/metabase overlays declared a column
just to attach descriptions. The duplicates carried no `type`, so the
pydantic SourceDefinition rejected them at semantic-query time and broke
`ktx sl query` for every overlay-backed measure. Now overlay columns
match base columns by name (case-insensitive): same-name entries merge
onto the manifest (overlay fields win, type/role fall back to the base,
descriptions merge per source key) and only new names append.
* refactor(sl): split overlay columns from column_overrides and enforce TS/Python wire contract
Overlay sources now have two distinct collections: `columns:` for computed
columns (requiring `expr` + `type`) and `column_overrides:` for metadata
patches to inherited manifest columns. Composing or loading an overlay that
mixes the two — or references an unknown column — fails with a typed error.
Introduce `ResolvedSemanticLayerSource` / `resolvedSourceSchema` /
`toResolvedWire` as the strict shape sent to the Python engine, and add a
schema contract test that diffs Zod against the Pydantic JSON schema dumped
by `python -m semantic_layer dump-schema`. `SourceDefinition` is now
`extra="forbid"` on the Python side.
`loadAllSources` surfaces per-file load errors instead of swallowing them,
so validation/query paths can report manifest shard parse failures.
* fix(context): make scan description generation resilient and quiet
A transient sampleTable failure during ingest used to take out every
table in a connection: generateTableDescription returned a hardcoded
'Table not found' string into descriptions.ai, and KtxDescriptionGenerator
was constructed without a logger, so the failure left no trail anywhere.
- sampleTable / sampleColumn calls retry 3x with 200/400/800ms backoff,
honouring KtxScanContext.signal via a new KtxAbortedError.
- On retry exhaustion or missing capability, table generation falls back
to a metadata-only prompt built from column name / native type / comment
/ rawDescriptions. The column path follows the same rule -- call the
LLM when any of samples or rawDescriptions are available; skip only
when both are absent.
- Logger is now threaded from KtxScanContext into the generator. Failures
emit structured KtxScanWarning entries (new description_fallback_used
code, plus existing sampling_failed / enrichment_failed /
connector_capability_missing). ktx scan groups warnings by code so a
batch of identical failures collapses to one summary line plus sample.
- Returns null on failure instead of the 'Table not found' sentinel; the
manifest writer's existing guard already skips empty descriptions, so
schema YAML no longer carries misleading text. SCAN_MANAGED_DESCRIPTION_KEYS
already strips stale 'ai' on merge, so existing YAML clears on next run.
Also suppress AI SDK v6 'system in messages' warning: pull system messages
out of KtxMessageBuilder.wrapSimple's output via a new splitKtxSystemMessages
helper and pass them top-level to generateText (preserves cacheControl
providerOptions on the SystemModelMessage). Agent-runner's local
splitSystemPromptMessages dedupes onto the shared helper.
* test(docs): align examples-docs assertions with revamped docs
PR #103 (setup/guide doc revamp) reworded several CLI examples and
connection labels; the assertions in scripts/examples-docs.test.mjs
still referenced the pre-revamp wording and were failing in CI on main.
Update the regexes to match the post-revamp content:
- drop the `--json` flag from the sl-query example expectation
- move the `Driver:` / `Status: ok` probe to the connection reference,
which is where that output now lives (driver id is lowercase
`postgres`, not the display name `PostgreSQL`)
- drop the obsolete `Install \`uv\`...` troubleshooting line
- accept `<connectionId>` everywhere; the docs no longer use the
hyphenated `<connection-id>` form
- match the `warehouse` connection id used in the quickstart instead of
the `postgres-warehouse` id only used in the README and setup ref
* fix(sl): skip TS/Python schema contract test when uv is unavailable
The TypeScript checks CI job does not install uv or Python, so the
module-level `execFileSync('uv', ...)` in schemas.contract.test.ts threw
ENOENT and failed the suite. Wrap the schema dump in a try/catch and
guard the describe block with `describe.skipIf` so the test skips in
environments without uv. Local dev and any CI job that has uv on PATH
still runs the cross-language contract assertion.
2026-05-15 02:11:04 +02:00
interface KtxDescriptionTableColumn {
name : string ;
nativeType? : string | null ;
comment? : string | null ;
}
2026-05-10 23:51:24 +02:00
export interface KtxDescriptionCachePort {
buildTableKey ( table : KtxTableRef ) : string ;
buildColumnKey ( table : KtxTableRef , columnName : string ) : string ;
2026-05-10 23:12:26 +02:00
buildConnectionKey ( connectionName : string ) : string ;
get ( key : string ) : Promise < string | null > ;
set ( key : string , value : string ) : Promise < void > ;
}
2026-05-10 23:51:24 +02:00
export interface KtxDescriptionSamplingPort {
2026-05-10 23:12:26 +02:00
id : string ;
2026-05-10 23:51:24 +02:00
sampleColumn ? ( input : KtxColumnSampleInput , ctx : KtxScanContext ) : Promise < KtxColumnSampleResult > ;
sampleTable ? ( input : KtxTableSampleInput , ctx : KtxScanContext ) : Promise < KtxTableSampleResult > ;
2026-05-10 23:12:26 +02:00
}
2026-05-10 23:51:24 +02:00
export interface KtxDescriptionGenerationSettings {
2026-05-10 23:12:26 +02:00
columnMaxWords : number ;
tableMaxWords : number ;
dataSourceMaxWords : number ;
temperature? : number ;
concurrencyLimit? : number ;
}
2026-05-10 23:51:24 +02:00
interface ResolvedKtxDescriptionGenerationSettings {
2026-05-10 23:12:26 +02:00
columnMaxWords : number ;
tableMaxWords : number ;
dataSourceMaxWords : number ;
temperature? : number ;
concurrencyLimit : number ;
}
2026-05-10 23:51:24 +02:00
export interface KtxDescriptionColumn {
2026-05-10 23:12:26 +02:00
name : string ;
type ? : string ;
rawDescriptions? : Record < string , string > ;
sampleValues? : unknown [ ] ;
}
2026-05-10 23:51:24 +02:00
export interface KtxDescriptionColumnTable extends KtxTableRef {
columns : KtxDescriptionColumn [ ] ;
2026-05-10 23:12:26 +02:00
}
2026-05-10 23:51:24 +02:00
export interface KtxDescriptionTableInput extends KtxTableRef {
2026-05-10 23:12:26 +02:00
rawDescriptions? : Record < string , string > ;
fix(context): merge overlay columns onto manifest columns by name (#94)
* fix(context): merge overlay columns onto manifest columns by name
composeOverlay was appending overlay columns to the manifest column list,
producing duplicate entries when dbt/metabase overlays declared a column
just to attach descriptions. The duplicates carried no `type`, so the
pydantic SourceDefinition rejected them at semantic-query time and broke
`ktx sl query` for every overlay-backed measure. Now overlay columns
match base columns by name (case-insensitive): same-name entries merge
onto the manifest (overlay fields win, type/role fall back to the base,
descriptions merge per source key) and only new names append.
* refactor(sl): split overlay columns from column_overrides and enforce TS/Python wire contract
Overlay sources now have two distinct collections: `columns:` for computed
columns (requiring `expr` + `type`) and `column_overrides:` for metadata
patches to inherited manifest columns. Composing or loading an overlay that
mixes the two — or references an unknown column — fails with a typed error.
Introduce `ResolvedSemanticLayerSource` / `resolvedSourceSchema` /
`toResolvedWire` as the strict shape sent to the Python engine, and add a
schema contract test that diffs Zod against the Pydantic JSON schema dumped
by `python -m semantic_layer dump-schema`. `SourceDefinition` is now
`extra="forbid"` on the Python side.
`loadAllSources` surfaces per-file load errors instead of swallowing them,
so validation/query paths can report manifest shard parse failures.
* fix(context): make scan description generation resilient and quiet
A transient sampleTable failure during ingest used to take out every
table in a connection: generateTableDescription returned a hardcoded
'Table not found' string into descriptions.ai, and KtxDescriptionGenerator
was constructed without a logger, so the failure left no trail anywhere.
- sampleTable / sampleColumn calls retry 3x with 200/400/800ms backoff,
honouring KtxScanContext.signal via a new KtxAbortedError.
- On retry exhaustion or missing capability, table generation falls back
to a metadata-only prompt built from column name / native type / comment
/ rawDescriptions. The column path follows the same rule -- call the
LLM when any of samples or rawDescriptions are available; skip only
when both are absent.
- Logger is now threaded from KtxScanContext into the generator. Failures
emit structured KtxScanWarning entries (new description_fallback_used
code, plus existing sampling_failed / enrichment_failed /
connector_capability_missing). ktx scan groups warnings by code so a
batch of identical failures collapses to one summary line plus sample.
- Returns null on failure instead of the 'Table not found' sentinel; the
manifest writer's existing guard already skips empty descriptions, so
schema YAML no longer carries misleading text. SCAN_MANAGED_DESCRIPTION_KEYS
already strips stale 'ai' on merge, so existing YAML clears on next run.
Also suppress AI SDK v6 'system in messages' warning: pull system messages
out of KtxMessageBuilder.wrapSimple's output via a new splitKtxSystemMessages
helper and pass them top-level to generateText (preserves cacheControl
providerOptions on the SystemModelMessage). Agent-runner's local
splitSystemPromptMessages dedupes onto the shared helper.
* test(docs): align examples-docs assertions with revamped docs
PR #103 (setup/guide doc revamp) reworded several CLI examples and
connection labels; the assertions in scripts/examples-docs.test.mjs
still referenced the pre-revamp wording and were failing in CI on main.
Update the regexes to match the post-revamp content:
- drop the `--json` flag from the sl-query example expectation
- move the `Driver:` / `Status: ok` probe to the connection reference,
which is where that output now lives (driver id is lowercase
`postgres`, not the display name `PostgreSQL`)
- drop the obsolete `Install \`uv\`...` troubleshooting line
- accept `<connectionId>` everywhere; the docs no longer use the
hyphenated `<connection-id>` form
- match the `warehouse` connection id used in the quickstart instead of
the `postgres-warehouse` id only used in the README and setup ref
* fix(sl): skip TS/Python schema contract test when uv is unavailable
The TypeScript checks CI job does not install uv or Python, so the
module-level `execFileSync('uv', ...)` in schemas.contract.test.ts threw
ENOENT and failed the suite. Wrap the schema dump in a try/catch and
guard the describe block with `describe.skipIf` so the test skips in
environments without uv. Local dev and any CI job that has uv on PATH
still runs the cross-language contract assertion.
2026-05-15 02:11:04 +02:00
columns? : KtxDescriptionTableColumn [ ] ;
2026-05-10 23:12:26 +02:00
}
2026-05-10 23:51:24 +02:00
export interface KtxColumnAnalysisResult {
2026-05-10 23:12:26 +02:00
columnDescriptions : Array < [ string , string | null ] > ;
processedColumns : string [ ] ;
skippedColumns : string [ ] ;
}
2026-05-10 23:51:24 +02:00
export interface KtxColumnDescriptionPromptInput {
2026-05-10 23:12:26 +02:00
columnName : string ;
columnValues : unknown [ ] ;
tableContext : string ;
dataSourceType : string ;
supportsNestedAnalysis : boolean ;
rawDescriptions? : Record < string , string > ;
}
2026-05-10 23:51:24 +02:00
export interface KtxTableDescriptionPromptInput {
2026-05-10 23:12:26 +02:00
tableName : string ;
fix(context): merge overlay columns onto manifest columns by name (#94)
* fix(context): merge overlay columns onto manifest columns by name
composeOverlay was appending overlay columns to the manifest column list,
producing duplicate entries when dbt/metabase overlays declared a column
just to attach descriptions. The duplicates carried no `type`, so the
pydantic SourceDefinition rejected them at semantic-query time and broke
`ktx sl query` for every overlay-backed measure. Now overlay columns
match base columns by name (case-insensitive): same-name entries merge
onto the manifest (overlay fields win, type/role fall back to the base,
descriptions merge per source key) and only new names append.
* refactor(sl): split overlay columns from column_overrides and enforce TS/Python wire contract
Overlay sources now have two distinct collections: `columns:` for computed
columns (requiring `expr` + `type`) and `column_overrides:` for metadata
patches to inherited manifest columns. Composing or loading an overlay that
mixes the two — or references an unknown column — fails with a typed error.
Introduce `ResolvedSemanticLayerSource` / `resolvedSourceSchema` /
`toResolvedWire` as the strict shape sent to the Python engine, and add a
schema contract test that diffs Zod against the Pydantic JSON schema dumped
by `python -m semantic_layer dump-schema`. `SourceDefinition` is now
`extra="forbid"` on the Python side.
`loadAllSources` surfaces per-file load errors instead of swallowing them,
so validation/query paths can report manifest shard parse failures.
* fix(context): make scan description generation resilient and quiet
A transient sampleTable failure during ingest used to take out every
table in a connection: generateTableDescription returned a hardcoded
'Table not found' string into descriptions.ai, and KtxDescriptionGenerator
was constructed without a logger, so the failure left no trail anywhere.
- sampleTable / sampleColumn calls retry 3x with 200/400/800ms backoff,
honouring KtxScanContext.signal via a new KtxAbortedError.
- On retry exhaustion or missing capability, table generation falls back
to a metadata-only prompt built from column name / native type / comment
/ rawDescriptions. The column path follows the same rule -- call the
LLM when any of samples or rawDescriptions are available; skip only
when both are absent.
- Logger is now threaded from KtxScanContext into the generator. Failures
emit structured KtxScanWarning entries (new description_fallback_used
code, plus existing sampling_failed / enrichment_failed /
connector_capability_missing). ktx scan groups warnings by code so a
batch of identical failures collapses to one summary line plus sample.
- Returns null on failure instead of the 'Table not found' sentinel; the
manifest writer's existing guard already skips empty descriptions, so
schema YAML no longer carries misleading text. SCAN_MANAGED_DESCRIPTION_KEYS
already strips stale 'ai' on merge, so existing YAML clears on next run.
Also suppress AI SDK v6 'system in messages' warning: pull system messages
out of KtxMessageBuilder.wrapSimple's output via a new splitKtxSystemMessages
helper and pass them top-level to generateText (preserves cacheControl
providerOptions on the SystemModelMessage). Agent-runner's local
splitSystemPromptMessages dedupes onto the shared helper.
* test(docs): align examples-docs assertions with revamped docs
PR #103 (setup/guide doc revamp) reworded several CLI examples and
connection labels; the assertions in scripts/examples-docs.test.mjs
still referenced the pre-revamp wording and were failing in CI on main.
Update the regexes to match the post-revamp content:
- drop the `--json` flag from the sl-query example expectation
- move the `Driver:` / `Status: ok` probe to the connection reference,
which is where that output now lives (driver id is lowercase
`postgres`, not the display name `PostgreSQL`)
- drop the obsolete `Install \`uv\`...` troubleshooting line
- accept `<connectionId>` everywhere; the docs no longer use the
hyphenated `<connection-id>` form
- match the `warehouse` connection id used in the quickstart instead of
the `postgres-warehouse` id only used in the README and setup ref
* fix(sl): skip TS/Python schema contract test when uv is unavailable
The TypeScript checks CI job does not install uv or Python, so the
module-level `execFileSync('uv', ...)` in schemas.contract.test.ts threw
ENOENT and failed the suite. Wrap the schema dump in a try/catch and
guard the describe block with `describe.skipIf` so the test skips in
environments without uv. Local dev and any CI job that has uv on PATH
still runs the cross-language contract assertion.
2026-05-15 02:11:04 +02:00
sampleData? : KtxTableSampleResult ;
columns? : KtxDescriptionTableColumn [ ] ;
2026-05-10 23:12:26 +02:00
dataSourceType : string ;
rawDescriptions? : Record < string , string > ;
}
2026-05-10 23:51:24 +02:00
export interface KtxDataSourceDescriptionPromptInput {
tableSamples : Array < [ string , KtxTableSampleResult ] > ;
2026-05-10 23:12:26 +02:00
dataSourceType : string ;
}
2026-05-10 23:51:24 +02:00
export interface KtxGenerateColumnDescriptionsInput {
2026-05-10 23:12:26 +02:00
connectionId : string ;
2026-05-10 23:51:24 +02:00
connector : KtxDescriptionSamplingPort ;
context : KtxScanContext ;
2026-05-10 23:12:26 +02:00
dataSourceType : string ;
supportsNestedAnalysis : boolean ;
2026-05-10 23:51:24 +02:00
table : KtxDescriptionColumnTable ;
2026-05-10 23:12:26 +02:00
skipExisting? : boolean ;
existingDescriptions? : Record < string , string | null > ;
}
2026-05-10 23:51:24 +02:00
export interface KtxGenerateTableDescriptionInput {
2026-05-10 23:12:26 +02:00
connectionId : string ;
2026-05-10 23:51:24 +02:00
connector : KtxDescriptionSamplingPort ;
context : KtxScanContext ;
2026-05-10 23:12:26 +02:00
dataSourceType : string ;
2026-05-10 23:51:24 +02:00
table : KtxDescriptionTableInput ;
2026-05-10 23:12:26 +02:00
}
2026-05-10 23:51:24 +02:00
export interface KtxGenerateDataSourceDescriptionInput {
2026-05-10 23:12:26 +02:00
connectionId : string ;
2026-05-10 23:51:24 +02:00
connector : KtxDescriptionSamplingPort ;
context : KtxScanContext ;
2026-05-10 23:12:26 +02:00
dataSourceType : string ;
2026-05-10 23:51:24 +02:00
tables : KtxTableRef [ ] ;
2026-05-10 23:12:26 +02:00
connectionName? : string ;
}
2026-05-10 23:51:24 +02:00
export interface KtxDescriptionGeneratorOptions {
llmProvider : KtxLlmProvider ;
cache? : KtxDescriptionCachePort ;
logger? : KtxScanLoggerPort ;
fix(context): merge overlay columns onto manifest columns by name (#94)
* fix(context): merge overlay columns onto manifest columns by name
composeOverlay was appending overlay columns to the manifest column list,
producing duplicate entries when dbt/metabase overlays declared a column
just to attach descriptions. The duplicates carried no `type`, so the
pydantic SourceDefinition rejected them at semantic-query time and broke
`ktx sl query` for every overlay-backed measure. Now overlay columns
match base columns by name (case-insensitive): same-name entries merge
onto the manifest (overlay fields win, type/role fall back to the base,
descriptions merge per source key) and only new names append.
* refactor(sl): split overlay columns from column_overrides and enforce TS/Python wire contract
Overlay sources now have two distinct collections: `columns:` for computed
columns (requiring `expr` + `type`) and `column_overrides:` for metadata
patches to inherited manifest columns. Composing or loading an overlay that
mixes the two — or references an unknown column — fails with a typed error.
Introduce `ResolvedSemanticLayerSource` / `resolvedSourceSchema` /
`toResolvedWire` as the strict shape sent to the Python engine, and add a
schema contract test that diffs Zod against the Pydantic JSON schema dumped
by `python -m semantic_layer dump-schema`. `SourceDefinition` is now
`extra="forbid"` on the Python side.
`loadAllSources` surfaces per-file load errors instead of swallowing them,
so validation/query paths can report manifest shard parse failures.
* fix(context): make scan description generation resilient and quiet
A transient sampleTable failure during ingest used to take out every
table in a connection: generateTableDescription returned a hardcoded
'Table not found' string into descriptions.ai, and KtxDescriptionGenerator
was constructed without a logger, so the failure left no trail anywhere.
- sampleTable / sampleColumn calls retry 3x with 200/400/800ms backoff,
honouring KtxScanContext.signal via a new KtxAbortedError.
- On retry exhaustion or missing capability, table generation falls back
to a metadata-only prompt built from column name / native type / comment
/ rawDescriptions. The column path follows the same rule -- call the
LLM when any of samples or rawDescriptions are available; skip only
when both are absent.
- Logger is now threaded from KtxScanContext into the generator. Failures
emit structured KtxScanWarning entries (new description_fallback_used
code, plus existing sampling_failed / enrichment_failed /
connector_capability_missing). ktx scan groups warnings by code so a
batch of identical failures collapses to one summary line plus sample.
- Returns null on failure instead of the 'Table not found' sentinel; the
manifest writer's existing guard already skips empty descriptions, so
schema YAML no longer carries misleading text. SCAN_MANAGED_DESCRIPTION_KEYS
already strips stale 'ai' on merge, so existing YAML clears on next run.
Also suppress AI SDK v6 'system in messages' warning: pull system messages
out of KtxMessageBuilder.wrapSimple's output via a new splitKtxSystemMessages
helper and pass them top-level to generateText (preserves cacheControl
providerOptions on the SystemModelMessage). Agent-runner's local
splitSystemPromptMessages dedupes onto the shared helper.
* test(docs): align examples-docs assertions with revamped docs
PR #103 (setup/guide doc revamp) reworded several CLI examples and
connection labels; the assertions in scripts/examples-docs.test.mjs
still referenced the pre-revamp wording and were failing in CI on main.
Update the regexes to match the post-revamp content:
- drop the `--json` flag from the sl-query example expectation
- move the `Driver:` / `Status: ok` probe to the connection reference,
which is where that output now lives (driver id is lowercase
`postgres`, not the display name `PostgreSQL`)
- drop the obsolete `Install \`uv\`...` troubleshooting line
- accept `<connectionId>` everywhere; the docs no longer use the
hyphenated `<connection-id>` form
- match the `warehouse` connection id used in the quickstart instead of
the `postgres-warehouse` id only used in the README and setup ref
* fix(sl): skip TS/Python schema contract test when uv is unavailable
The TypeScript checks CI job does not install uv or Python, so the
module-level `execFileSync('uv', ...)` in schemas.contract.test.ts threw
ENOENT and failed the suite. Wrap the schema dump in a try/catch and
guard the describe block with `describe.skipIf` so the test skips in
environments without uv. Local dev and any CI job that has uv on PATH
still runs the cross-language contract assertion.
2026-05-15 02:11:04 +02:00
onWarning ? : ( warning : KtxScanWarning ) = > void ;
2026-05-10 23:51:24 +02:00
settings : KtxDescriptionGenerationSettings ;
2026-05-10 23:12:26 +02:00
}
interface ColumnTaskResult {
columnName : string ;
description : string | null ;
processed : boolean ;
skipped : boolean ;
}
function descriptionSources ( rawDescriptions : Record < string , string > | undefined ) : Array < [ string , string ] > {
if ( ! rawDescriptions ) {
return [ ] ;
}
return Object . entries ( rawDescriptions ) . filter ( ( [ source , text ] ) = > source !== 'ai' && source !== 'user' && ! ! text ) ;
}
function errorMessage ( error : unknown ) : string {
return error instanceof Error ? error.message : String ( error ) ;
}
fix(context): merge overlay columns onto manifest columns by name (#94)
* fix(context): merge overlay columns onto manifest columns by name
composeOverlay was appending overlay columns to the manifest column list,
producing duplicate entries when dbt/metabase overlays declared a column
just to attach descriptions. The duplicates carried no `type`, so the
pydantic SourceDefinition rejected them at semantic-query time and broke
`ktx sl query` for every overlay-backed measure. Now overlay columns
match base columns by name (case-insensitive): same-name entries merge
onto the manifest (overlay fields win, type/role fall back to the base,
descriptions merge per source key) and only new names append.
* refactor(sl): split overlay columns from column_overrides and enforce TS/Python wire contract
Overlay sources now have two distinct collections: `columns:` for computed
columns (requiring `expr` + `type`) and `column_overrides:` for metadata
patches to inherited manifest columns. Composing or loading an overlay that
mixes the two — or references an unknown column — fails with a typed error.
Introduce `ResolvedSemanticLayerSource` / `resolvedSourceSchema` /
`toResolvedWire` as the strict shape sent to the Python engine, and add a
schema contract test that diffs Zod against the Pydantic JSON schema dumped
by `python -m semantic_layer dump-schema`. `SourceDefinition` is now
`extra="forbid"` on the Python side.
`loadAllSources` surfaces per-file load errors instead of swallowing them,
so validation/query paths can report manifest shard parse failures.
* fix(context): make scan description generation resilient and quiet
A transient sampleTable failure during ingest used to take out every
table in a connection: generateTableDescription returned a hardcoded
'Table not found' string into descriptions.ai, and KtxDescriptionGenerator
was constructed without a logger, so the failure left no trail anywhere.
- sampleTable / sampleColumn calls retry 3x with 200/400/800ms backoff,
honouring KtxScanContext.signal via a new KtxAbortedError.
- On retry exhaustion or missing capability, table generation falls back
to a metadata-only prompt built from column name / native type / comment
/ rawDescriptions. The column path follows the same rule -- call the
LLM when any of samples or rawDescriptions are available; skip only
when both are absent.
- Logger is now threaded from KtxScanContext into the generator. Failures
emit structured KtxScanWarning entries (new description_fallback_used
code, plus existing sampling_failed / enrichment_failed /
connector_capability_missing). ktx scan groups warnings by code so a
batch of identical failures collapses to one summary line plus sample.
- Returns null on failure instead of the 'Table not found' sentinel; the
manifest writer's existing guard already skips empty descriptions, so
schema YAML no longer carries misleading text. SCAN_MANAGED_DESCRIPTION_KEYS
already strips stale 'ai' on merge, so existing YAML clears on next run.
Also suppress AI SDK v6 'system in messages' warning: pull system messages
out of KtxMessageBuilder.wrapSimple's output via a new splitKtxSystemMessages
helper and pass them top-level to generateText (preserves cacheControl
providerOptions on the SystemModelMessage). Agent-runner's local
splitSystemPromptMessages dedupes onto the shared helper.
* test(docs): align examples-docs assertions with revamped docs
PR #103 (setup/guide doc revamp) reworded several CLI examples and
connection labels; the assertions in scripts/examples-docs.test.mjs
still referenced the pre-revamp wording and were failing in CI on main.
Update the regexes to match the post-revamp content:
- drop the `--json` flag from the sl-query example expectation
- move the `Driver:` / `Status: ok` probe to the connection reference,
which is where that output now lives (driver id is lowercase
`postgres`, not the display name `PostgreSQL`)
- drop the obsolete `Install \`uv\`...` troubleshooting line
- accept `<connectionId>` everywhere; the docs no longer use the
hyphenated `<connection-id>` form
- match the `warehouse` connection id used in the quickstart instead of
the `postgres-warehouse` id only used in the README and setup ref
* fix(sl): skip TS/Python schema contract test when uv is unavailable
The TypeScript checks CI job does not install uv or Python, so the
module-level `execFileSync('uv', ...)` in schemas.contract.test.ts threw
ENOENT and failed the suite. Wrap the schema dump in a try/catch and
guard the describe block with `describe.skipIf` so the test skips in
environments without uv. Local dev and any CI job that has uv on PATH
still runs the cross-language contract assertion.
2026-05-15 02:11:04 +02:00
class KtxAbortedError extends Error {
constructor ( ) {
super ( 'aborted' ) ;
this . name = 'KtxAbortedError' ;
}
}
async function delayWithAbort ( ms : number , signal? : AbortSignal ) : Promise < void > {
if ( ! signal ) {
await new Promise < void > ( ( resolve ) = > setTimeout ( resolve , ms ) ) ;
return ;
}
if ( signal . aborted ) {
throw new KtxAbortedError ( ) ;
}
await new Promise < void > ( ( resolve , reject ) = > {
const timer = setTimeout ( ( ) = > {
signal . removeEventListener ( 'abort' , onAbort ) ;
resolve ( ) ;
} , ms ) ;
const onAbort = ( ) : void = > {
clearTimeout ( timer ) ;
reject ( new KtxAbortedError ( ) ) ;
} ;
signal . addEventListener ( 'abort' , onAbort , { once : true } ) ;
} ) ;
}
interface RetryAsyncOptions {
attempts : number ;
baseDelayMs : number ;
signal? : AbortSignal ;
onAttemptFailure ? : ( error : unknown , attempt : number ) = > void ;
}
async function retryAsync < T > ( fn : ( ) = > Promise < T > , options : RetryAsyncOptions ) : Promise < T > {
const attempts = Math . max ( 1 , options . attempts ) ;
let lastError : unknown ;
for ( let attempt = 1 ; attempt <= attempts ; attempt += 1 ) {
if ( options . signal ? . aborted ) {
throw new KtxAbortedError ( ) ;
}
try {
return await fn ( ) ;
} catch ( error ) {
lastError = error ;
if ( error instanceof KtxAbortedError ) {
throw error ;
}
options . onAttemptFailure ? . ( error , attempt ) ;
if ( attempt === attempts ) {
break ;
}
const delay = options . baseDelayMs * 2 * * ( attempt - 1 ) ;
await delayWithAbort ( delay , options . signal ) ;
}
}
throw lastError ;
}
2026-05-10 23:51:24 +02:00
function toTableRef ( table : KtxTableRef ) : KtxTableRef {
2026-05-10 23:12:26 +02:00
return {
catalog : table.catalog ,
db : table.db ,
name : table.name ,
} ;
}
async function runWithConcurrency < TInput , TOutput > (
items : readonly TInput [ ] ,
concurrencyLimit : number ,
worker : ( item : TInput , index : number ) = > Promise < TOutput > ,
) : Promise < TOutput [ ] > {
const results : TOutput [ ] = [ ] ;
let nextIndex = 0 ;
const workerCount = Math . max ( 1 , Math . min ( concurrencyLimit , items . length || 1 ) ) ;
await Promise . all (
Array . from ( { length : workerCount } , async ( ) = > {
while ( nextIndex < items . length ) {
const index = nextIndex ;
nextIndex += 1 ;
const item = items [ index ] ;
if ( item !== undefined ) {
results [ index ] = await worker ( item , index ) ;
}
}
} ) ,
) ;
return results ;
}
2026-05-14 15:36:27 +02:00
export interface KtxDescriptionPrompt {
system : string ;
user : string ;
2026-05-10 23:12:26 +02:00
}
2026-05-14 15:36:27 +02:00
function wordLimitLine ( maxWords : number ) : string {
return ` Please provide a concise description in ${ maxWords } words or less. ` ;
}
export function buildKtxColumnDescriptionPrompt (
input : KtxColumnDescriptionPromptInput & { maxWords? : number } ,
) : KtxDescriptionPrompt {
2026-05-10 23:12:26 +02:00
const sampleValues = input . columnValues . slice ( 0 , 5 ) ;
const valuesStr = sampleValues
. filter ( ( value ) = > value !== null && value !== undefined )
. map ( ( value ) = > String ( value ) )
. join ( ', ' ) ;
2026-05-14 15:36:27 +02:00
const systemParts : string [ ] = [
` Analyze database columns and provide a concise description.
Provide a brief description of what the column contains without repeating the column name .
Focus on the data ' s meaning and business purpose . Start directly with the content description .
Example :
"first names of individuals, likely employees or contacts" instead of "The column contains first names..."
"Job titles or roles of individuals..." instead of "This column contains job titles..." ` ,
] ;
if ( input . dataSourceType === 'BIGQUERY' && input . supportsNestedAnalysis ) {
systemParts . push (
'If the sampled values indicate nested/structured data (JSON, STRUCT, or ARRAY), describe its general business purpose and data organization.' ,
) ;
}
if ( input . maxWords !== undefined ) {
systemParts . push ( wordLimitLine ( input . maxWords ) ) ;
}
2026-05-10 23:12:26 +02:00
fix(context): merge overlay columns onto manifest columns by name (#94)
* fix(context): merge overlay columns onto manifest columns by name
composeOverlay was appending overlay columns to the manifest column list,
producing duplicate entries when dbt/metabase overlays declared a column
just to attach descriptions. The duplicates carried no `type`, so the
pydantic SourceDefinition rejected them at semantic-query time and broke
`ktx sl query` for every overlay-backed measure. Now overlay columns
match base columns by name (case-insensitive): same-name entries merge
onto the manifest (overlay fields win, type/role fall back to the base,
descriptions merge per source key) and only new names append.
* refactor(sl): split overlay columns from column_overrides and enforce TS/Python wire contract
Overlay sources now have two distinct collections: `columns:` for computed
columns (requiring `expr` + `type`) and `column_overrides:` for metadata
patches to inherited manifest columns. Composing or loading an overlay that
mixes the two — or references an unknown column — fails with a typed error.
Introduce `ResolvedSemanticLayerSource` / `resolvedSourceSchema` /
`toResolvedWire` as the strict shape sent to the Python engine, and add a
schema contract test that diffs Zod against the Pydantic JSON schema dumped
by `python -m semantic_layer dump-schema`. `SourceDefinition` is now
`extra="forbid"` on the Python side.
`loadAllSources` surfaces per-file load errors instead of swallowing them,
so validation/query paths can report manifest shard parse failures.
* fix(context): make scan description generation resilient and quiet
A transient sampleTable failure during ingest used to take out every
table in a connection: generateTableDescription returned a hardcoded
'Table not found' string into descriptions.ai, and KtxDescriptionGenerator
was constructed without a logger, so the failure left no trail anywhere.
- sampleTable / sampleColumn calls retry 3x with 200/400/800ms backoff,
honouring KtxScanContext.signal via a new KtxAbortedError.
- On retry exhaustion or missing capability, table generation falls back
to a metadata-only prompt built from column name / native type / comment
/ rawDescriptions. The column path follows the same rule -- call the
LLM when any of samples or rawDescriptions are available; skip only
when both are absent.
- Logger is now threaded from KtxScanContext into the generator. Failures
emit structured KtxScanWarning entries (new description_fallback_used
code, plus existing sampling_failed / enrichment_failed /
connector_capability_missing). ktx scan groups warnings by code so a
batch of identical failures collapses to one summary line plus sample.
- Returns null on failure instead of the 'Table not found' sentinel; the
manifest writer's existing guard already skips empty descriptions, so
schema YAML no longer carries misleading text. SCAN_MANAGED_DESCRIPTION_KEYS
already strips stale 'ai' on merge, so existing YAML clears on next run.
Also suppress AI SDK v6 'system in messages' warning: pull system messages
out of KtxMessageBuilder.wrapSimple's output via a new splitKtxSystemMessages
helper and pass them top-level to generateText (preserves cacheControl
providerOptions on the SystemModelMessage). Agent-runner's local
splitSystemPromptMessages dedupes onto the shared helper.
* test(docs): align examples-docs assertions with revamped docs
PR #103 (setup/guide doc revamp) reworded several CLI examples and
connection labels; the assertions in scripts/examples-docs.test.mjs
still referenced the pre-revamp wording and were failing in CI on main.
Update the regexes to match the post-revamp content:
- drop the `--json` flag from the sl-query example expectation
- move the `Driver:` / `Status: ok` probe to the connection reference,
which is where that output now lives (driver id is lowercase
`postgres`, not the display name `PostgreSQL`)
- drop the obsolete `Install \`uv\`...` troubleshooting line
- accept `<connectionId>` everywhere; the docs no longer use the
hyphenated `<connection-id>` form
- match the `warehouse` connection id used in the quickstart instead of
the `postgres-warehouse` id only used in the README and setup ref
* fix(sl): skip TS/Python schema contract test when uv is unavailable
The TypeScript checks CI job does not install uv or Python, so the
module-level `execFileSync('uv', ...)` in schemas.contract.test.ts threw
ENOENT and failed the suite. Wrap the schema dump in a try/catch and
guard the describe block with `describe.skipIf` so the test skips in
environments without uv. Local dev and any CI job that has uv on PATH
still runs the cross-language contract assertion.
2026-05-15 02:11:04 +02:00
const sampleValuesContent = valuesStr . length > 0 ? valuesStr : 'unavailable' ;
2026-05-14 15:36:27 +02:00
let user = ` <table_context> ${ input . tableContext } </table_context>
2026-05-10 23:12:26 +02:00
< column_name > $ { input . columnName } < / column_name >
fix(context): merge overlay columns onto manifest columns by name (#94)
* fix(context): merge overlay columns onto manifest columns by name
composeOverlay was appending overlay columns to the manifest column list,
producing duplicate entries when dbt/metabase overlays declared a column
just to attach descriptions. The duplicates carried no `type`, so the
pydantic SourceDefinition rejected them at semantic-query time and broke
`ktx sl query` for every overlay-backed measure. Now overlay columns
match base columns by name (case-insensitive): same-name entries merge
onto the manifest (overlay fields win, type/role fall back to the base,
descriptions merge per source key) and only new names append.
* refactor(sl): split overlay columns from column_overrides and enforce TS/Python wire contract
Overlay sources now have two distinct collections: `columns:` for computed
columns (requiring `expr` + `type`) and `column_overrides:` for metadata
patches to inherited manifest columns. Composing or loading an overlay that
mixes the two — or references an unknown column — fails with a typed error.
Introduce `ResolvedSemanticLayerSource` / `resolvedSourceSchema` /
`toResolvedWire` as the strict shape sent to the Python engine, and add a
schema contract test that diffs Zod against the Pydantic JSON schema dumped
by `python -m semantic_layer dump-schema`. `SourceDefinition` is now
`extra="forbid"` on the Python side.
`loadAllSources` surfaces per-file load errors instead of swallowing them,
so validation/query paths can report manifest shard parse failures.
* fix(context): make scan description generation resilient and quiet
A transient sampleTable failure during ingest used to take out every
table in a connection: generateTableDescription returned a hardcoded
'Table not found' string into descriptions.ai, and KtxDescriptionGenerator
was constructed without a logger, so the failure left no trail anywhere.
- sampleTable / sampleColumn calls retry 3x with 200/400/800ms backoff,
honouring KtxScanContext.signal via a new KtxAbortedError.
- On retry exhaustion or missing capability, table generation falls back
to a metadata-only prompt built from column name / native type / comment
/ rawDescriptions. The column path follows the same rule -- call the
LLM when any of samples or rawDescriptions are available; skip only
when both are absent.
- Logger is now threaded from KtxScanContext into the generator. Failures
emit structured KtxScanWarning entries (new description_fallback_used
code, plus existing sampling_failed / enrichment_failed /
connector_capability_missing). ktx scan groups warnings by code so a
batch of identical failures collapses to one summary line plus sample.
- Returns null on failure instead of the 'Table not found' sentinel; the
manifest writer's existing guard already skips empty descriptions, so
schema YAML no longer carries misleading text. SCAN_MANAGED_DESCRIPTION_KEYS
already strips stale 'ai' on merge, so existing YAML clears on next run.
Also suppress AI SDK v6 'system in messages' warning: pull system messages
out of KtxMessageBuilder.wrapSimple's output via a new splitKtxSystemMessages
helper and pass them top-level to generateText (preserves cacheControl
providerOptions on the SystemModelMessage). Agent-runner's local
splitSystemPromptMessages dedupes onto the shared helper.
* test(docs): align examples-docs assertions with revamped docs
PR #103 (setup/guide doc revamp) reworded several CLI examples and
connection labels; the assertions in scripts/examples-docs.test.mjs
still referenced the pre-revamp wording and were failing in CI on main.
Update the regexes to match the post-revamp content:
- drop the `--json` flag from the sl-query example expectation
- move the `Driver:` / `Status: ok` probe to the connection reference,
which is where that output now lives (driver id is lowercase
`postgres`, not the display name `PostgreSQL`)
- drop the obsolete `Install \`uv\`...` troubleshooting line
- accept `<connectionId>` everywhere; the docs no longer use the
hyphenated `<connection-id>` form
- match the `warehouse` connection id used in the quickstart instead of
the `postgres-warehouse` id only used in the README and setup ref
* fix(sl): skip TS/Python schema contract test when uv is unavailable
The TypeScript checks CI job does not install uv or Python, so the
module-level `execFileSync('uv', ...)` in schemas.contract.test.ts threw
ENOENT and failed the suite. Wrap the schema dump in a try/catch and
guard the describe block with `describe.skipIf` so the test skips in
environments without uv. Local dev and any CI job that has uv on PATH
still runs the cross-language contract assertion.
2026-05-15 02:11:04 +02:00
< sample_values > $ { sampleValuesContent } < / sample_values >
2026-05-10 23:12:26 +02:00
` ;
const sources = descriptionSources ( input . rawDescriptions ) ;
if ( sources . length > 0 ) {
2026-05-14 15:36:27 +02:00
user += '\nExisting descriptions from other sources:\n' ;
2026-05-10 23:12:26 +02:00
for ( const [ source , text ] of sources ) {
2026-05-14 15:36:27 +02:00
user += ` < ${ source } _documentation> ${ text } </ ${ source } _documentation> \ n ` ;
2026-05-10 23:12:26 +02:00
}
2026-05-14 15:36:27 +02:00
user +=
2026-05-10 23:12:26 +02:00
'\nSynthesize a description that captures the most important information from all sources. Prioritize the sources as authoritative context.\n' ;
}
2026-05-14 15:36:27 +02:00
return { system : systemParts.join ( '\n\n' ) , user : user.trim ( ) } ;
2026-05-10 23:12:26 +02:00
}
2026-05-14 15:36:27 +02:00
export function buildKtxTableDescriptionPrompt (
input : KtxTableDescriptionPromptInput & { maxWords? : number } ,
) : KtxDescriptionPrompt {
const systemParts : string [ ] = [
` Analyze database tables and provide a concise description.
2026-05-10 23:12:26 +02:00
2026-05-14 15:36:27 +02:00
Provide a brief description of what the table represents and its business purpose .
Do NOT list or describe individual columns or fields .
Start directly with the content description without mentioning the table name .
Focus on the data ' s meaning and business purpose .
Example : "Information about healthcare professionals used for workforce management" instead of "The blahblah table contains information about healthcare professionals including their names, titles..." ` ,
] ;
if ( input . dataSourceType === 'BIGQUERY' ) {
systemParts . push (
"Note (don't include in the final answer): BigQuery tables may contain nested structures, arrays, or other complex data types." ,
) ;
}
if ( input . maxWords !== undefined ) {
systemParts . push ( wordLimitLine ( input . maxWords ) ) ;
}
fix(context): merge overlay columns onto manifest columns by name (#94)
* fix(context): merge overlay columns onto manifest columns by name
composeOverlay was appending overlay columns to the manifest column list,
producing duplicate entries when dbt/metabase overlays declared a column
just to attach descriptions. The duplicates carried no `type`, so the
pydantic SourceDefinition rejected them at semantic-query time and broke
`ktx sl query` for every overlay-backed measure. Now overlay columns
match base columns by name (case-insensitive): same-name entries merge
onto the manifest (overlay fields win, type/role fall back to the base,
descriptions merge per source key) and only new names append.
* refactor(sl): split overlay columns from column_overrides and enforce TS/Python wire contract
Overlay sources now have two distinct collections: `columns:` for computed
columns (requiring `expr` + `type`) and `column_overrides:` for metadata
patches to inherited manifest columns. Composing or loading an overlay that
mixes the two — or references an unknown column — fails with a typed error.
Introduce `ResolvedSemanticLayerSource` / `resolvedSourceSchema` /
`toResolvedWire` as the strict shape sent to the Python engine, and add a
schema contract test that diffs Zod against the Pydantic JSON schema dumped
by `python -m semantic_layer dump-schema`. `SourceDefinition` is now
`extra="forbid"` on the Python side.
`loadAllSources` surfaces per-file load errors instead of swallowing them,
so validation/query paths can report manifest shard parse failures.
* fix(context): make scan description generation resilient and quiet
A transient sampleTable failure during ingest used to take out every
table in a connection: generateTableDescription returned a hardcoded
'Table not found' string into descriptions.ai, and KtxDescriptionGenerator
was constructed without a logger, so the failure left no trail anywhere.
- sampleTable / sampleColumn calls retry 3x with 200/400/800ms backoff,
honouring KtxScanContext.signal via a new KtxAbortedError.
- On retry exhaustion or missing capability, table generation falls back
to a metadata-only prompt built from column name / native type / comment
/ rawDescriptions. The column path follows the same rule -- call the
LLM when any of samples or rawDescriptions are available; skip only
when both are absent.
- Logger is now threaded from KtxScanContext into the generator. Failures
emit structured KtxScanWarning entries (new description_fallback_used
code, plus existing sampling_failed / enrichment_failed /
connector_capability_missing). ktx scan groups warnings by code so a
batch of identical failures collapses to one summary line plus sample.
- Returns null on failure instead of the 'Table not found' sentinel; the
manifest writer's existing guard already skips empty descriptions, so
schema YAML no longer carries misleading text. SCAN_MANAGED_DESCRIPTION_KEYS
already strips stale 'ai' on merge, so existing YAML clears on next run.
Also suppress AI SDK v6 'system in messages' warning: pull system messages
out of KtxMessageBuilder.wrapSimple's output via a new splitKtxSystemMessages
helper and pass them top-level to generateText (preserves cacheControl
providerOptions on the SystemModelMessage). Agent-runner's local
splitSystemPromptMessages dedupes onto the shared helper.
* test(docs): align examples-docs assertions with revamped docs
PR #103 (setup/guide doc revamp) reworded several CLI examples and
connection labels; the assertions in scripts/examples-docs.test.mjs
still referenced the pre-revamp wording and were failing in CI on main.
Update the regexes to match the post-revamp content:
- drop the `--json` flag from the sl-query example expectation
- move the `Driver:` / `Status: ok` probe to the connection reference,
which is where that output now lives (driver id is lowercase
`postgres`, not the display name `PostgreSQL`)
- drop the obsolete `Install \`uv\`...` troubleshooting line
- accept `<connectionId>` everywhere; the docs no longer use the
hyphenated `<connection-id>` form
- match the `warehouse` connection id used in the quickstart instead of
the `postgres-warehouse` id only used in the README and setup ref
* fix(sl): skip TS/Python schema contract test when uv is unavailable
The TypeScript checks CI job does not install uv or Python, so the
module-level `execFileSync('uv', ...)` in schemas.contract.test.ts threw
ENOENT and failed the suite. Wrap the schema dump in a try/catch and
guard the describe block with `describe.skipIf` so the test skips in
environments without uv. Local dev and any CI job that has uv on PATH
still runs the cross-language contract assertion.
2026-05-15 02:11:04 +02:00
const hasSamples = ! ! input . sampleData && input . sampleData . rows . length > 0 ;
let columnsLine : string ;
let rowsLine : string ;
if ( hasSamples ) {
const sampleData = input . sampleData ! ;
const columnInfo : string [ ] = [ ] ;
for ( let index = 0 ; index < Math . min ( sampleData . headers . length , 10 ) ; index += 1 ) {
const header = sampleData . headers [ index ] ;
const sampleValues = sampleData . rows
. slice ( 0 , 3 )
. map ( ( row ) = > row [ index ] )
. filter ( ( value ) = > value !== null && value !== undefined ) ;
columnInfo . push ( ` ${ header } : ${ sampleValues . map ( ( value ) = > String ( value ) ) . join ( ', ' ) } ` ) ;
}
columnsLine = ` Columns and sample data: ${ columnInfo . join ( ' | ' ) } ` ;
rowsLine = ` Total rows in sample: ${ sampleData . rows . length } ` ;
} else if ( input . columns && input . columns . length > 0 ) {
const columnInfo = input . columns . slice ( 0 , 30 ) . map ( ( column ) = > {
const typePart = column . nativeType ? ` ( ${ column . nativeType } ) ` : '' ;
const commentPart = column . comment ? ` — ${ column . comment } ` : '' ;
return ` ${ column . name } ${ typePart } ${ commentPart } ` ;
} ) ;
columnsLine = ` Columns (metadata only, no sample rows): ${ columnInfo . join ( ' | ' ) } ` ;
rowsLine = 'Sample rows: unavailable' ;
} else {
columnsLine = 'Columns: unavailable' ;
rowsLine = 'Sample rows: unavailable' ;
}
2026-05-14 15:36:27 +02:00
let user = ` Table: ${ input . tableName }
fix(context): merge overlay columns onto manifest columns by name (#94)
* fix(context): merge overlay columns onto manifest columns by name
composeOverlay was appending overlay columns to the manifest column list,
producing duplicate entries when dbt/metabase overlays declared a column
just to attach descriptions. The duplicates carried no `type`, so the
pydantic SourceDefinition rejected them at semantic-query time and broke
`ktx sl query` for every overlay-backed measure. Now overlay columns
match base columns by name (case-insensitive): same-name entries merge
onto the manifest (overlay fields win, type/role fall back to the base,
descriptions merge per source key) and only new names append.
* refactor(sl): split overlay columns from column_overrides and enforce TS/Python wire contract
Overlay sources now have two distinct collections: `columns:` for computed
columns (requiring `expr` + `type`) and `column_overrides:` for metadata
patches to inherited manifest columns. Composing or loading an overlay that
mixes the two — or references an unknown column — fails with a typed error.
Introduce `ResolvedSemanticLayerSource` / `resolvedSourceSchema` /
`toResolvedWire` as the strict shape sent to the Python engine, and add a
schema contract test that diffs Zod against the Pydantic JSON schema dumped
by `python -m semantic_layer dump-schema`. `SourceDefinition` is now
`extra="forbid"` on the Python side.
`loadAllSources` surfaces per-file load errors instead of swallowing them,
so validation/query paths can report manifest shard parse failures.
* fix(context): make scan description generation resilient and quiet
A transient sampleTable failure during ingest used to take out every
table in a connection: generateTableDescription returned a hardcoded
'Table not found' string into descriptions.ai, and KtxDescriptionGenerator
was constructed without a logger, so the failure left no trail anywhere.
- sampleTable / sampleColumn calls retry 3x with 200/400/800ms backoff,
honouring KtxScanContext.signal via a new KtxAbortedError.
- On retry exhaustion or missing capability, table generation falls back
to a metadata-only prompt built from column name / native type / comment
/ rawDescriptions. The column path follows the same rule -- call the
LLM when any of samples or rawDescriptions are available; skip only
when both are absent.
- Logger is now threaded from KtxScanContext into the generator. Failures
emit structured KtxScanWarning entries (new description_fallback_used
code, plus existing sampling_failed / enrichment_failed /
connector_capability_missing). ktx scan groups warnings by code so a
batch of identical failures collapses to one summary line plus sample.
- Returns null on failure instead of the 'Table not found' sentinel; the
manifest writer's existing guard already skips empty descriptions, so
schema YAML no longer carries misleading text. SCAN_MANAGED_DESCRIPTION_KEYS
already strips stale 'ai' on merge, so existing YAML clears on next run.
Also suppress AI SDK v6 'system in messages' warning: pull system messages
out of KtxMessageBuilder.wrapSimple's output via a new splitKtxSystemMessages
helper and pass them top-level to generateText (preserves cacheControl
providerOptions on the SystemModelMessage). Agent-runner's local
splitSystemPromptMessages dedupes onto the shared helper.
* test(docs): align examples-docs assertions with revamped docs
PR #103 (setup/guide doc revamp) reworded several CLI examples and
connection labels; the assertions in scripts/examples-docs.test.mjs
still referenced the pre-revamp wording and were failing in CI on main.
Update the regexes to match the post-revamp content:
- drop the `--json` flag from the sl-query example expectation
- move the `Driver:` / `Status: ok` probe to the connection reference,
which is where that output now lives (driver id is lowercase
`postgres`, not the display name `PostgreSQL`)
- drop the obsolete `Install \`uv\`...` troubleshooting line
- accept `<connectionId>` everywhere; the docs no longer use the
hyphenated `<connection-id>` form
- match the `warehouse` connection id used in the quickstart instead of
the `postgres-warehouse` id only used in the README and setup ref
* fix(sl): skip TS/Python schema contract test when uv is unavailable
The TypeScript checks CI job does not install uv or Python, so the
module-level `execFileSync('uv', ...)` in schemas.contract.test.ts threw
ENOENT and failed the suite. Wrap the schema dump in a try/catch and
guard the describe block with `describe.skipIf` so the test skips in
environments without uv. Local dev and any CI job that has uv on PATH
still runs the cross-language contract assertion.
2026-05-15 02:11:04 +02:00
$ { columnsLine }
$ { rowsLine }
2026-05-14 15:36:27 +02:00
Data source type : $ { input . dataSourceType } ` ;
2026-05-10 23:12:26 +02:00
const sources = descriptionSources ( input . rawDescriptions ) ;
if ( sources . length > 0 ) {
2026-05-14 15:36:27 +02:00
user += '\n\nExisting descriptions from other sources:\n' ;
2026-05-10 23:12:26 +02:00
for ( const [ source , text ] of sources ) {
2026-05-14 15:36:27 +02:00
user += ` ${ source } : ${ text } \ n ` ;
2026-05-10 23:12:26 +02:00
}
2026-05-14 15:36:27 +02:00
user +=
'\nSynthesize a description that captures the most important information from all sources. Prioritize the sources as authoritative context.' ;
2026-05-10 23:12:26 +02:00
}
2026-05-14 15:36:27 +02:00
return { system : systemParts.join ( '\n\n' ) , user : user.trim ( ) } ;
2026-05-10 23:12:26 +02:00
}
2026-05-14 15:36:27 +02:00
export function buildKtxDataSourceDescriptionPrompt (
input : KtxDataSourceDescriptionPromptInput & { maxWords? : number } ,
) : KtxDescriptionPrompt {
2026-05-10 23:12:26 +02:00
const tablesText = input . tableSamples
. map (
( [ tableName , sampleData ] ) = >
` ${ tableName } ( ${ sampleData . headers . length } columns, ${ sampleData . rows . length } sample rows) ` ,
)
. join ( ' | ' ) ;
2026-05-14 15:36:27 +02:00
const systemParts : string [ ] = [
` Analyze databases and provide a concise description.
2026-05-10 23:12:26 +02:00
2026-05-14 15:36:27 +02:00
Provide a direct , concise description of what the database represents and its business purpose .
Do NOT start with phrases like "This database appears to represent" or "This BigQuery dataset" .
Start directly with the domain or business area description .
Focus on the overall data model and its intended use .
Example : "Healthcare-related database with a focus on patient management..." instead of "This database appears to represent a healthcare-related system..." ` ,
] ;
2026-05-10 23:12:26 +02:00
if ( input . dataSourceType === 'BIGQUERY' ) {
2026-05-14 15:36:27 +02:00
systemParts . push (
"Note (don't include in the final answer): BigQuery datasets may contain large-scale analytics data, nested structures, and complex data types." ,
) ;
}
if ( input . maxWords !== undefined ) {
systemParts . push ( wordLimitLine ( input . maxWords ) ) ;
2026-05-10 23:12:26 +02:00
}
2026-05-14 15:36:27 +02:00
const user = ` Tables: ${ tablesText }
Total tables analyzed : $ { input . tableSamples . length }
Data source type : $ { input . dataSourceType } ` ;
2026-05-10 23:12:26 +02:00
2026-05-14 15:36:27 +02:00
return { system : systemParts.join ( '\n\n' ) , user } ;
2026-05-10 23:12:26 +02:00
}
2026-05-10 23:51:24 +02:00
export class KtxDescriptionGenerator {
private readonly llmProvider : KtxLlmProvider ;
private readonly cache? : KtxDescriptionCachePort ;
private readonly logger? : KtxScanLoggerPort ;
fix(context): merge overlay columns onto manifest columns by name (#94)
* fix(context): merge overlay columns onto manifest columns by name
composeOverlay was appending overlay columns to the manifest column list,
producing duplicate entries when dbt/metabase overlays declared a column
just to attach descriptions. The duplicates carried no `type`, so the
pydantic SourceDefinition rejected them at semantic-query time and broke
`ktx sl query` for every overlay-backed measure. Now overlay columns
match base columns by name (case-insensitive): same-name entries merge
onto the manifest (overlay fields win, type/role fall back to the base,
descriptions merge per source key) and only new names append.
* refactor(sl): split overlay columns from column_overrides and enforce TS/Python wire contract
Overlay sources now have two distinct collections: `columns:` for computed
columns (requiring `expr` + `type`) and `column_overrides:` for metadata
patches to inherited manifest columns. Composing or loading an overlay that
mixes the two — or references an unknown column — fails with a typed error.
Introduce `ResolvedSemanticLayerSource` / `resolvedSourceSchema` /
`toResolvedWire` as the strict shape sent to the Python engine, and add a
schema contract test that diffs Zod against the Pydantic JSON schema dumped
by `python -m semantic_layer dump-schema`. `SourceDefinition` is now
`extra="forbid"` on the Python side.
`loadAllSources` surfaces per-file load errors instead of swallowing them,
so validation/query paths can report manifest shard parse failures.
* fix(context): make scan description generation resilient and quiet
A transient sampleTable failure during ingest used to take out every
table in a connection: generateTableDescription returned a hardcoded
'Table not found' string into descriptions.ai, and KtxDescriptionGenerator
was constructed without a logger, so the failure left no trail anywhere.
- sampleTable / sampleColumn calls retry 3x with 200/400/800ms backoff,
honouring KtxScanContext.signal via a new KtxAbortedError.
- On retry exhaustion or missing capability, table generation falls back
to a metadata-only prompt built from column name / native type / comment
/ rawDescriptions. The column path follows the same rule -- call the
LLM when any of samples or rawDescriptions are available; skip only
when both are absent.
- Logger is now threaded from KtxScanContext into the generator. Failures
emit structured KtxScanWarning entries (new description_fallback_used
code, plus existing sampling_failed / enrichment_failed /
connector_capability_missing). ktx scan groups warnings by code so a
batch of identical failures collapses to one summary line plus sample.
- Returns null on failure instead of the 'Table not found' sentinel; the
manifest writer's existing guard already skips empty descriptions, so
schema YAML no longer carries misleading text. SCAN_MANAGED_DESCRIPTION_KEYS
already strips stale 'ai' on merge, so existing YAML clears on next run.
Also suppress AI SDK v6 'system in messages' warning: pull system messages
out of KtxMessageBuilder.wrapSimple's output via a new splitKtxSystemMessages
helper and pass them top-level to generateText (preserves cacheControl
providerOptions on the SystemModelMessage). Agent-runner's local
splitSystemPromptMessages dedupes onto the shared helper.
* test(docs): align examples-docs assertions with revamped docs
PR #103 (setup/guide doc revamp) reworded several CLI examples and
connection labels; the assertions in scripts/examples-docs.test.mjs
still referenced the pre-revamp wording and were failing in CI on main.
Update the regexes to match the post-revamp content:
- drop the `--json` flag from the sl-query example expectation
- move the `Driver:` / `Status: ok` probe to the connection reference,
which is where that output now lives (driver id is lowercase
`postgres`, not the display name `PostgreSQL`)
- drop the obsolete `Install \`uv\`...` troubleshooting line
- accept `<connectionId>` everywhere; the docs no longer use the
hyphenated `<connection-id>` form
- match the `warehouse` connection id used in the quickstart instead of
the `postgres-warehouse` id only used in the README and setup ref
* fix(sl): skip TS/Python schema contract test when uv is unavailable
The TypeScript checks CI job does not install uv or Python, so the
module-level `execFileSync('uv', ...)` in schemas.contract.test.ts threw
ENOENT and failed the suite. Wrap the schema dump in a try/catch and
guard the describe block with `describe.skipIf` so the test skips in
environments without uv. Local dev and any CI job that has uv on PATH
still runs the cross-language contract assertion.
2026-05-15 02:11:04 +02:00
private readonly onWarning ? : ( warning : KtxScanWarning ) = > void ;
2026-05-10 23:51:24 +02:00
private readonly settings : ResolvedKtxDescriptionGenerationSettings ;
2026-05-10 23:12:26 +02:00
2026-05-10 23:51:24 +02:00
constructor ( options : KtxDescriptionGeneratorOptions ) {
2026-05-10 23:12:26 +02:00
this . llmProvider = options . llmProvider ;
this . cache = options . cache ;
this . logger = options . logger ;
fix(context): merge overlay columns onto manifest columns by name (#94)
* fix(context): merge overlay columns onto manifest columns by name
composeOverlay was appending overlay columns to the manifest column list,
producing duplicate entries when dbt/metabase overlays declared a column
just to attach descriptions. The duplicates carried no `type`, so the
pydantic SourceDefinition rejected them at semantic-query time and broke
`ktx sl query` for every overlay-backed measure. Now overlay columns
match base columns by name (case-insensitive): same-name entries merge
onto the manifest (overlay fields win, type/role fall back to the base,
descriptions merge per source key) and only new names append.
* refactor(sl): split overlay columns from column_overrides and enforce TS/Python wire contract
Overlay sources now have two distinct collections: `columns:` for computed
columns (requiring `expr` + `type`) and `column_overrides:` for metadata
patches to inherited manifest columns. Composing or loading an overlay that
mixes the two — or references an unknown column — fails with a typed error.
Introduce `ResolvedSemanticLayerSource` / `resolvedSourceSchema` /
`toResolvedWire` as the strict shape sent to the Python engine, and add a
schema contract test that diffs Zod against the Pydantic JSON schema dumped
by `python -m semantic_layer dump-schema`. `SourceDefinition` is now
`extra="forbid"` on the Python side.
`loadAllSources` surfaces per-file load errors instead of swallowing them,
so validation/query paths can report manifest shard parse failures.
* fix(context): make scan description generation resilient and quiet
A transient sampleTable failure during ingest used to take out every
table in a connection: generateTableDescription returned a hardcoded
'Table not found' string into descriptions.ai, and KtxDescriptionGenerator
was constructed without a logger, so the failure left no trail anywhere.
- sampleTable / sampleColumn calls retry 3x with 200/400/800ms backoff,
honouring KtxScanContext.signal via a new KtxAbortedError.
- On retry exhaustion or missing capability, table generation falls back
to a metadata-only prompt built from column name / native type / comment
/ rawDescriptions. The column path follows the same rule -- call the
LLM when any of samples or rawDescriptions are available; skip only
when both are absent.
- Logger is now threaded from KtxScanContext into the generator. Failures
emit structured KtxScanWarning entries (new description_fallback_used
code, plus existing sampling_failed / enrichment_failed /
connector_capability_missing). ktx scan groups warnings by code so a
batch of identical failures collapses to one summary line plus sample.
- Returns null on failure instead of the 'Table not found' sentinel; the
manifest writer's existing guard already skips empty descriptions, so
schema YAML no longer carries misleading text. SCAN_MANAGED_DESCRIPTION_KEYS
already strips stale 'ai' on merge, so existing YAML clears on next run.
Also suppress AI SDK v6 'system in messages' warning: pull system messages
out of KtxMessageBuilder.wrapSimple's output via a new splitKtxSystemMessages
helper and pass them top-level to generateText (preserves cacheControl
providerOptions on the SystemModelMessage). Agent-runner's local
splitSystemPromptMessages dedupes onto the shared helper.
* test(docs): align examples-docs assertions with revamped docs
PR #103 (setup/guide doc revamp) reworded several CLI examples and
connection labels; the assertions in scripts/examples-docs.test.mjs
still referenced the pre-revamp wording and were failing in CI on main.
Update the regexes to match the post-revamp content:
- drop the `--json` flag from the sl-query example expectation
- move the `Driver:` / `Status: ok` probe to the connection reference,
which is where that output now lives (driver id is lowercase
`postgres`, not the display name `PostgreSQL`)
- drop the obsolete `Install \`uv\`...` troubleshooting line
- accept `<connectionId>` everywhere; the docs no longer use the
hyphenated `<connection-id>` form
- match the `warehouse` connection id used in the quickstart instead of
the `postgres-warehouse` id only used in the README and setup ref
* fix(sl): skip TS/Python schema contract test when uv is unavailable
The TypeScript checks CI job does not install uv or Python, so the
module-level `execFileSync('uv', ...)` in schemas.contract.test.ts threw
ENOENT and failed the suite. Wrap the schema dump in a try/catch and
guard the describe block with `describe.skipIf` so the test skips in
environments without uv. Local dev and any CI job that has uv on PATH
still runs the cross-language contract assertion.
2026-05-15 02:11:04 +02:00
this . onWarning = options . onWarning ;
2026-05-10 23:12:26 +02:00
this . settings = {
columnMaxWords : options.settings.columnMaxWords ,
tableMaxWords : options.settings.tableMaxWords ,
dataSourceMaxWords : options.settings.dataSourceMaxWords ,
. . . ( options . settings . temperature !== undefined ? { temperature : options.settings.temperature } : { } ) ,
concurrencyLimit : options.settings.concurrencyLimit ? ? 5 ,
} ;
}
2026-05-10 23:51:24 +02:00
async generateColumnDescriptions ( input : KtxGenerateColumnDescriptionsInput ) : Promise < KtxColumnAnalysisResult > {
2026-05-10 23:12:26 +02:00
const columnsToProcess = input . table . columns ;
const tableContext = ` Table: ${ input . table . name } | Columns: ${ columnsToProcess . map ( ( column ) = > column . name ) . join ( ', ' ) } | Data source: ${ input . dataSourceType } ` ;
const results = await runWithConcurrency ( columnsToProcess , this . settings . concurrencyLimit , async ( column ) = >
this . generateOneColumnDescription ( input , column , tableContext ) ,
) ;
const columnDescriptions : Array < [ string , string | null ] > = [ ] ;
const processedColumns : string [ ] = [ ] ;
const skippedColumns : string [ ] = [ ] ;
for ( const result of results ) {
columnDescriptions . push ( [ result . columnName , result . description ] ) ;
if ( result . skipped ) {
skippedColumns . push ( result . columnName ) ;
} else if ( result . processed ) {
processedColumns . push ( result . columnName ) ;
}
}
return {
columnDescriptions ,
processedColumns ,
skippedColumns ,
} ;
}
2026-05-12 14:34:15 +02:00
async generateTableDescription ( input : KtxGenerateTableDescriptionInput ) : Promise < string | null > {
2026-05-10 23:12:26 +02:00
const tableRef = toTableRef ( input . table ) ;
const cacheKey = this . cache ? . buildTableKey ( tableRef ) ;
if ( cacheKey ) {
const cached = await this . cache ? . get ( cacheKey ) ;
if ( cached ) {
return cached ;
}
}
fix(context): merge overlay columns onto manifest columns by name (#94)
* fix(context): merge overlay columns onto manifest columns by name
composeOverlay was appending overlay columns to the manifest column list,
producing duplicate entries when dbt/metabase overlays declared a column
just to attach descriptions. The duplicates carried no `type`, so the
pydantic SourceDefinition rejected them at semantic-query time and broke
`ktx sl query` for every overlay-backed measure. Now overlay columns
match base columns by name (case-insensitive): same-name entries merge
onto the manifest (overlay fields win, type/role fall back to the base,
descriptions merge per source key) and only new names append.
* refactor(sl): split overlay columns from column_overrides and enforce TS/Python wire contract
Overlay sources now have two distinct collections: `columns:` for computed
columns (requiring `expr` + `type`) and `column_overrides:` for metadata
patches to inherited manifest columns. Composing or loading an overlay that
mixes the two — or references an unknown column — fails with a typed error.
Introduce `ResolvedSemanticLayerSource` / `resolvedSourceSchema` /
`toResolvedWire` as the strict shape sent to the Python engine, and add a
schema contract test that diffs Zod against the Pydantic JSON schema dumped
by `python -m semantic_layer dump-schema`. `SourceDefinition` is now
`extra="forbid"` on the Python side.
`loadAllSources` surfaces per-file load errors instead of swallowing them,
so validation/query paths can report manifest shard parse failures.
* fix(context): make scan description generation resilient and quiet
A transient sampleTable failure during ingest used to take out every
table in a connection: generateTableDescription returned a hardcoded
'Table not found' string into descriptions.ai, and KtxDescriptionGenerator
was constructed without a logger, so the failure left no trail anywhere.
- sampleTable / sampleColumn calls retry 3x with 200/400/800ms backoff,
honouring KtxScanContext.signal via a new KtxAbortedError.
- On retry exhaustion or missing capability, table generation falls back
to a metadata-only prompt built from column name / native type / comment
/ rawDescriptions. The column path follows the same rule -- call the
LLM when any of samples or rawDescriptions are available; skip only
when both are absent.
- Logger is now threaded from KtxScanContext into the generator. Failures
emit structured KtxScanWarning entries (new description_fallback_used
code, plus existing sampling_failed / enrichment_failed /
connector_capability_missing). ktx scan groups warnings by code so a
batch of identical failures collapses to one summary line plus sample.
- Returns null on failure instead of the 'Table not found' sentinel; the
manifest writer's existing guard already skips empty descriptions, so
schema YAML no longer carries misleading text. SCAN_MANAGED_DESCRIPTION_KEYS
already strips stale 'ai' on merge, so existing YAML clears on next run.
Also suppress AI SDK v6 'system in messages' warning: pull system messages
out of KtxMessageBuilder.wrapSimple's output via a new splitKtxSystemMessages
helper and pass them top-level to generateText (preserves cacheControl
providerOptions on the SystemModelMessage). Agent-runner's local
splitSystemPromptMessages dedupes onto the shared helper.
* test(docs): align examples-docs assertions with revamped docs
PR #103 (setup/guide doc revamp) reworded several CLI examples and
connection labels; the assertions in scripts/examples-docs.test.mjs
still referenced the pre-revamp wording and were failing in CI on main.
Update the regexes to match the post-revamp content:
- drop the `--json` flag from the sl-query example expectation
- move the `Driver:` / `Status: ok` probe to the connection reference,
which is where that output now lives (driver id is lowercase
`postgres`, not the display name `PostgreSQL`)
- drop the obsolete `Install \`uv\`...` troubleshooting line
- accept `<connectionId>` everywhere; the docs no longer use the
hyphenated `<connection-id>` form
- match the `warehouse` connection id used in the quickstart instead of
the `postgres-warehouse` id only used in the README and setup ref
* fix(sl): skip TS/Python schema contract test when uv is unavailable
The TypeScript checks CI job does not install uv or Python, so the
module-level `execFileSync('uv', ...)` in schemas.contract.test.ts threw
ENOENT and failed the suite. Wrap the schema dump in a try/catch and
guard the describe block with `describe.skipIf` so the test skips in
environments without uv. Local dev and any CI job that has uv on PATH
still runs the cross-language contract assertion.
2026-05-15 02:11:04 +02:00
const sampleTable = input . connector . sampleTable ;
let sampleData : KtxTableSampleResult | null = null ;
let fallbackReason : 'capability_missing' | 'sampling_failed' | 'empty_sample' | null = null ;
if ( ! sampleTable ) {
fallbackReason = 'capability_missing' ;
this . logger ? . warn ( 'KTX scan connector does not support table sampling; falling back to metadata-only prompt' , {
2026-05-10 23:12:26 +02:00
connectorId : input.connector.id ,
table : input.table.name ,
} ) ;
fix(context): merge overlay columns onto manifest columns by name (#94)
* fix(context): merge overlay columns onto manifest columns by name
composeOverlay was appending overlay columns to the manifest column list,
producing duplicate entries when dbt/metabase overlays declared a column
just to attach descriptions. The duplicates carried no `type`, so the
pydantic SourceDefinition rejected them at semantic-query time and broke
`ktx sl query` for every overlay-backed measure. Now overlay columns
match base columns by name (case-insensitive): same-name entries merge
onto the manifest (overlay fields win, type/role fall back to the base,
descriptions merge per source key) and only new names append.
* refactor(sl): split overlay columns from column_overrides and enforce TS/Python wire contract
Overlay sources now have two distinct collections: `columns:` for computed
columns (requiring `expr` + `type`) and `column_overrides:` for metadata
patches to inherited manifest columns. Composing or loading an overlay that
mixes the two — or references an unknown column — fails with a typed error.
Introduce `ResolvedSemanticLayerSource` / `resolvedSourceSchema` /
`toResolvedWire` as the strict shape sent to the Python engine, and add a
schema contract test that diffs Zod against the Pydantic JSON schema dumped
by `python -m semantic_layer dump-schema`. `SourceDefinition` is now
`extra="forbid"` on the Python side.
`loadAllSources` surfaces per-file load errors instead of swallowing them,
so validation/query paths can report manifest shard parse failures.
* fix(context): make scan description generation resilient and quiet
A transient sampleTable failure during ingest used to take out every
table in a connection: generateTableDescription returned a hardcoded
'Table not found' string into descriptions.ai, and KtxDescriptionGenerator
was constructed without a logger, so the failure left no trail anywhere.
- sampleTable / sampleColumn calls retry 3x with 200/400/800ms backoff,
honouring KtxScanContext.signal via a new KtxAbortedError.
- On retry exhaustion or missing capability, table generation falls back
to a metadata-only prompt built from column name / native type / comment
/ rawDescriptions. The column path follows the same rule -- call the
LLM when any of samples or rawDescriptions are available; skip only
when both are absent.
- Logger is now threaded from KtxScanContext into the generator. Failures
emit structured KtxScanWarning entries (new description_fallback_used
code, plus existing sampling_failed / enrichment_failed /
connector_capability_missing). ktx scan groups warnings by code so a
batch of identical failures collapses to one summary line plus sample.
- Returns null on failure instead of the 'Table not found' sentinel; the
manifest writer's existing guard already skips empty descriptions, so
schema YAML no longer carries misleading text. SCAN_MANAGED_DESCRIPTION_KEYS
already strips stale 'ai' on merge, so existing YAML clears on next run.
Also suppress AI SDK v6 'system in messages' warning: pull system messages
out of KtxMessageBuilder.wrapSimple's output via a new splitKtxSystemMessages
helper and pass them top-level to generateText (preserves cacheControl
providerOptions on the SystemModelMessage). Agent-runner's local
splitSystemPromptMessages dedupes onto the shared helper.
* test(docs): align examples-docs assertions with revamped docs
PR #103 (setup/guide doc revamp) reworded several CLI examples and
connection labels; the assertions in scripts/examples-docs.test.mjs
still referenced the pre-revamp wording and were failing in CI on main.
Update the regexes to match the post-revamp content:
- drop the `--json` flag from the sl-query example expectation
- move the `Driver:` / `Status: ok` probe to the connection reference,
which is where that output now lives (driver id is lowercase
`postgres`, not the display name `PostgreSQL`)
- drop the obsolete `Install \`uv\`...` troubleshooting line
- accept `<connectionId>` everywhere; the docs no longer use the
hyphenated `<connection-id>` form
- match the `warehouse` connection id used in the quickstart instead of
the `postgres-warehouse` id only used in the README and setup ref
* fix(sl): skip TS/Python schema contract test when uv is unavailable
The TypeScript checks CI job does not install uv or Python, so the
module-level `execFileSync('uv', ...)` in schemas.contract.test.ts threw
ENOENT and failed the suite. Wrap the schema dump in a try/catch and
guard the describe block with `describe.skipIf` so the test skips in
environments without uv. Local dev and any CI job that has uv on PATH
still runs the cross-language contract assertion.
2026-05-15 02:11:04 +02:00
this . onWarning ? . ( {
code : 'connector_capability_missing' ,
message : ` Connector ${ input . connector . id } does not support sampleTable; using metadata-only description prompt ` ,
table : input.table.name ,
recoverable : true ,
metadata : { connectorId : input.connector.id , capability : 'sampleTable' } ,
} ) ;
} else {
try {
sampleData = await retryAsync (
( ) = >
sampleTable (
{
connectionId : input.connectionId ,
table : tableRef ,
limit : 20 ,
} ,
input . context ,
) ,
{
attempts : 3 ,
baseDelayMs : 200 ,
signal : input.context.signal ,
onAttemptFailure : ( error , attempt ) = > {
this . logger ? . warn (
` sampleTable attempt ${ attempt } failed for ${ input . table . name } : ${ errorMessage ( error ) } ` ,
{
connectorId : input.connector.id ,
table : input.table.name ,
attempt ,
} ,
) ;
} ,
} ,
) ;
if ( sampleData . rows . length === 0 ) {
fallbackReason = 'empty_sample' ;
this . logger ? . warn ( 'sampleTable returned no rows; using metadata-only prompt' , {
connectorId : input.connector.id ,
table : input.table.name ,
} ) ;
}
} catch ( error ) {
if ( error instanceof KtxAbortedError ) {
throw error ;
}
fallbackReason = 'sampling_failed' ;
this . logger ? . error ( ` sampleTable exhausted retries for ${ input . table . name } : ${ errorMessage ( error ) } ` , {
connectorId : input.connector.id ,
table : input.table.name ,
} ) ;
this . onWarning ? . ( {
code : 'sampling_failed' ,
message : ` Failed to sample table ${ input . table . name } after retries: ${ errorMessage ( error ) } ` ,
table : input.table.name ,
recoverable : true ,
metadata : { connectorId : input.connector.id , error : errorMessage ( error ) } ,
} ) ;
}
2026-05-10 23:12:26 +02:00
}
try {
2026-05-10 23:51:24 +02:00
const prompt = buildKtxTableDescriptionPrompt ( {
2026-05-10 23:12:26 +02:00
tableName : input.table.name ,
fix(context): merge overlay columns onto manifest columns by name (#94)
* fix(context): merge overlay columns onto manifest columns by name
composeOverlay was appending overlay columns to the manifest column list,
producing duplicate entries when dbt/metabase overlays declared a column
just to attach descriptions. The duplicates carried no `type`, so the
pydantic SourceDefinition rejected them at semantic-query time and broke
`ktx sl query` for every overlay-backed measure. Now overlay columns
match base columns by name (case-insensitive): same-name entries merge
onto the manifest (overlay fields win, type/role fall back to the base,
descriptions merge per source key) and only new names append.
* refactor(sl): split overlay columns from column_overrides and enforce TS/Python wire contract
Overlay sources now have two distinct collections: `columns:` for computed
columns (requiring `expr` + `type`) and `column_overrides:` for metadata
patches to inherited manifest columns. Composing or loading an overlay that
mixes the two — or references an unknown column — fails with a typed error.
Introduce `ResolvedSemanticLayerSource` / `resolvedSourceSchema` /
`toResolvedWire` as the strict shape sent to the Python engine, and add a
schema contract test that diffs Zod against the Pydantic JSON schema dumped
by `python -m semantic_layer dump-schema`. `SourceDefinition` is now
`extra="forbid"` on the Python side.
`loadAllSources` surfaces per-file load errors instead of swallowing them,
so validation/query paths can report manifest shard parse failures.
* fix(context): make scan description generation resilient and quiet
A transient sampleTable failure during ingest used to take out every
table in a connection: generateTableDescription returned a hardcoded
'Table not found' string into descriptions.ai, and KtxDescriptionGenerator
was constructed without a logger, so the failure left no trail anywhere.
- sampleTable / sampleColumn calls retry 3x with 200/400/800ms backoff,
honouring KtxScanContext.signal via a new KtxAbortedError.
- On retry exhaustion or missing capability, table generation falls back
to a metadata-only prompt built from column name / native type / comment
/ rawDescriptions. The column path follows the same rule -- call the
LLM when any of samples or rawDescriptions are available; skip only
when both are absent.
- Logger is now threaded from KtxScanContext into the generator. Failures
emit structured KtxScanWarning entries (new description_fallback_used
code, plus existing sampling_failed / enrichment_failed /
connector_capability_missing). ktx scan groups warnings by code so a
batch of identical failures collapses to one summary line plus sample.
- Returns null on failure instead of the 'Table not found' sentinel; the
manifest writer's existing guard already skips empty descriptions, so
schema YAML no longer carries misleading text. SCAN_MANAGED_DESCRIPTION_KEYS
already strips stale 'ai' on merge, so existing YAML clears on next run.
Also suppress AI SDK v6 'system in messages' warning: pull system messages
out of KtxMessageBuilder.wrapSimple's output via a new splitKtxSystemMessages
helper and pass them top-level to generateText (preserves cacheControl
providerOptions on the SystemModelMessage). Agent-runner's local
splitSystemPromptMessages dedupes onto the shared helper.
* test(docs): align examples-docs assertions with revamped docs
PR #103 (setup/guide doc revamp) reworded several CLI examples and
connection labels; the assertions in scripts/examples-docs.test.mjs
still referenced the pre-revamp wording and were failing in CI on main.
Update the regexes to match the post-revamp content:
- drop the `--json` flag from the sl-query example expectation
- move the `Driver:` / `Status: ok` probe to the connection reference,
which is where that output now lives (driver id is lowercase
`postgres`, not the display name `PostgreSQL`)
- drop the obsolete `Install \`uv\`...` troubleshooting line
- accept `<connectionId>` everywhere; the docs no longer use the
hyphenated `<connection-id>` form
- match the `warehouse` connection id used in the quickstart instead of
the `postgres-warehouse` id only used in the README and setup ref
* fix(sl): skip TS/Python schema contract test when uv is unavailable
The TypeScript checks CI job does not install uv or Python, so the
module-level `execFileSync('uv', ...)` in schemas.contract.test.ts threw
ENOENT and failed the suite. Wrap the schema dump in a try/catch and
guard the describe block with `describe.skipIf` so the test skips in
environments without uv. Local dev and any CI job that has uv on PATH
still runs the cross-language contract assertion.
2026-05-15 02:11:04 +02:00
. . . ( fallbackReason === null && sampleData ? { sampleData } : { } ) ,
. . . ( input . table . columns && input . table . columns . length > 0 ? { columns : input.table.columns } : { } ) ,
2026-05-10 23:12:26 +02:00
dataSourceType : input.dataSourceType ,
rawDescriptions : input.table.rawDescriptions ,
2026-05-14 15:36:27 +02:00
maxWords : this.settings.tableMaxWords ,
2026-05-10 23:12:26 +02:00
} ) ;
2026-05-14 15:36:27 +02:00
const description = await this . generateAiDescription ( prompt , 'ktx-table-description' ) ;
2026-05-12 14:34:15 +02:00
if ( cacheKey && description ) {
2026-05-10 23:12:26 +02:00
await this . cache ? . set ( cacheKey , description ) ;
}
fix(context): merge overlay columns onto manifest columns by name (#94)
* fix(context): merge overlay columns onto manifest columns by name
composeOverlay was appending overlay columns to the manifest column list,
producing duplicate entries when dbt/metabase overlays declared a column
just to attach descriptions. The duplicates carried no `type`, so the
pydantic SourceDefinition rejected them at semantic-query time and broke
`ktx sl query` for every overlay-backed measure. Now overlay columns
match base columns by name (case-insensitive): same-name entries merge
onto the manifest (overlay fields win, type/role fall back to the base,
descriptions merge per source key) and only new names append.
* refactor(sl): split overlay columns from column_overrides and enforce TS/Python wire contract
Overlay sources now have two distinct collections: `columns:` for computed
columns (requiring `expr` + `type`) and `column_overrides:` for metadata
patches to inherited manifest columns. Composing or loading an overlay that
mixes the two — or references an unknown column — fails with a typed error.
Introduce `ResolvedSemanticLayerSource` / `resolvedSourceSchema` /
`toResolvedWire` as the strict shape sent to the Python engine, and add a
schema contract test that diffs Zod against the Pydantic JSON schema dumped
by `python -m semantic_layer dump-schema`. `SourceDefinition` is now
`extra="forbid"` on the Python side.
`loadAllSources` surfaces per-file load errors instead of swallowing them,
so validation/query paths can report manifest shard parse failures.
* fix(context): make scan description generation resilient and quiet
A transient sampleTable failure during ingest used to take out every
table in a connection: generateTableDescription returned a hardcoded
'Table not found' string into descriptions.ai, and KtxDescriptionGenerator
was constructed without a logger, so the failure left no trail anywhere.
- sampleTable / sampleColumn calls retry 3x with 200/400/800ms backoff,
honouring KtxScanContext.signal via a new KtxAbortedError.
- On retry exhaustion or missing capability, table generation falls back
to a metadata-only prompt built from column name / native type / comment
/ rawDescriptions. The column path follows the same rule -- call the
LLM when any of samples or rawDescriptions are available; skip only
when both are absent.
- Logger is now threaded from KtxScanContext into the generator. Failures
emit structured KtxScanWarning entries (new description_fallback_used
code, plus existing sampling_failed / enrichment_failed /
connector_capability_missing). ktx scan groups warnings by code so a
batch of identical failures collapses to one summary line plus sample.
- Returns null on failure instead of the 'Table not found' sentinel; the
manifest writer's existing guard already skips empty descriptions, so
schema YAML no longer carries misleading text. SCAN_MANAGED_DESCRIPTION_KEYS
already strips stale 'ai' on merge, so existing YAML clears on next run.
Also suppress AI SDK v6 'system in messages' warning: pull system messages
out of KtxMessageBuilder.wrapSimple's output via a new splitKtxSystemMessages
helper and pass them top-level to generateText (preserves cacheControl
providerOptions on the SystemModelMessage). Agent-runner's local
splitSystemPromptMessages dedupes onto the shared helper.
* test(docs): align examples-docs assertions with revamped docs
PR #103 (setup/guide doc revamp) reworded several CLI examples and
connection labels; the assertions in scripts/examples-docs.test.mjs
still referenced the pre-revamp wording and were failing in CI on main.
Update the regexes to match the post-revamp content:
- drop the `--json` flag from the sl-query example expectation
- move the `Driver:` / `Status: ok` probe to the connection reference,
which is where that output now lives (driver id is lowercase
`postgres`, not the display name `PostgreSQL`)
- drop the obsolete `Install \`uv\`...` troubleshooting line
- accept `<connectionId>` everywhere; the docs no longer use the
hyphenated `<connection-id>` form
- match the `warehouse` connection id used in the quickstart instead of
the `postgres-warehouse` id only used in the README and setup ref
* fix(sl): skip TS/Python schema contract test when uv is unavailable
The TypeScript checks CI job does not install uv or Python, so the
module-level `execFileSync('uv', ...)` in schemas.contract.test.ts threw
ENOENT and failed the suite. Wrap the schema dump in a try/catch and
guard the describe block with `describe.skipIf` so the test skips in
environments without uv. Local dev and any CI job that has uv on PATH
still runs the cross-language contract assertion.
2026-05-15 02:11:04 +02:00
if ( description && fallbackReason !== null ) {
this . onWarning ? . ( {
code : 'description_fallback_used' ,
message : ` Generated table description without sample rows for ${ input . table . name } (reason: ${ fallbackReason } ) ` ,
table : input.table.name ,
recoverable : true ,
metadata : { connectorId : input.connector.id , reason : fallbackReason } ,
} ) ;
}
if ( ! description ) {
this . onWarning ? . ( {
code : 'enrichment_failed' ,
message : ` Failed to generate description for table ${ input . table . name } ` ,
table : input.table.name ,
recoverable : true ,
metadata : { connectorId : input.connector.id , usedFallback : fallbackReason !== null } ,
} ) ;
}
2026-05-10 23:12:26 +02:00
return description ;
} catch ( error ) {
fix(context): merge overlay columns onto manifest columns by name (#94)
* fix(context): merge overlay columns onto manifest columns by name
composeOverlay was appending overlay columns to the manifest column list,
producing duplicate entries when dbt/metabase overlays declared a column
just to attach descriptions. The duplicates carried no `type`, so the
pydantic SourceDefinition rejected them at semantic-query time and broke
`ktx sl query` for every overlay-backed measure. Now overlay columns
match base columns by name (case-insensitive): same-name entries merge
onto the manifest (overlay fields win, type/role fall back to the base,
descriptions merge per source key) and only new names append.
* refactor(sl): split overlay columns from column_overrides and enforce TS/Python wire contract
Overlay sources now have two distinct collections: `columns:` for computed
columns (requiring `expr` + `type`) and `column_overrides:` for metadata
patches to inherited manifest columns. Composing or loading an overlay that
mixes the two — or references an unknown column — fails with a typed error.
Introduce `ResolvedSemanticLayerSource` / `resolvedSourceSchema` /
`toResolvedWire` as the strict shape sent to the Python engine, and add a
schema contract test that diffs Zod against the Pydantic JSON schema dumped
by `python -m semantic_layer dump-schema`. `SourceDefinition` is now
`extra="forbid"` on the Python side.
`loadAllSources` surfaces per-file load errors instead of swallowing them,
so validation/query paths can report manifest shard parse failures.
* fix(context): make scan description generation resilient and quiet
A transient sampleTable failure during ingest used to take out every
table in a connection: generateTableDescription returned a hardcoded
'Table not found' string into descriptions.ai, and KtxDescriptionGenerator
was constructed without a logger, so the failure left no trail anywhere.
- sampleTable / sampleColumn calls retry 3x with 200/400/800ms backoff,
honouring KtxScanContext.signal via a new KtxAbortedError.
- On retry exhaustion or missing capability, table generation falls back
to a metadata-only prompt built from column name / native type / comment
/ rawDescriptions. The column path follows the same rule -- call the
LLM when any of samples or rawDescriptions are available; skip only
when both are absent.
- Logger is now threaded from KtxScanContext into the generator. Failures
emit structured KtxScanWarning entries (new description_fallback_used
code, plus existing sampling_failed / enrichment_failed /
connector_capability_missing). ktx scan groups warnings by code so a
batch of identical failures collapses to one summary line plus sample.
- Returns null on failure instead of the 'Table not found' sentinel; the
manifest writer's existing guard already skips empty descriptions, so
schema YAML no longer carries misleading text. SCAN_MANAGED_DESCRIPTION_KEYS
already strips stale 'ai' on merge, so existing YAML clears on next run.
Also suppress AI SDK v6 'system in messages' warning: pull system messages
out of KtxMessageBuilder.wrapSimple's output via a new splitKtxSystemMessages
helper and pass them top-level to generateText (preserves cacheControl
providerOptions on the SystemModelMessage). Agent-runner's local
splitSystemPromptMessages dedupes onto the shared helper.
* test(docs): align examples-docs assertions with revamped docs
PR #103 (setup/guide doc revamp) reworded several CLI examples and
connection labels; the assertions in scripts/examples-docs.test.mjs
still referenced the pre-revamp wording and were failing in CI on main.
Update the regexes to match the post-revamp content:
- drop the `--json` flag from the sl-query example expectation
- move the `Driver:` / `Status: ok` probe to the connection reference,
which is where that output now lives (driver id is lowercase
`postgres`, not the display name `PostgreSQL`)
- drop the obsolete `Install \`uv\`...` troubleshooting line
- accept `<connectionId>` everywhere; the docs no longer use the
hyphenated `<connection-id>` form
- match the `warehouse` connection id used in the quickstart instead of
the `postgres-warehouse` id only used in the README and setup ref
* fix(sl): skip TS/Python schema contract test when uv is unavailable
The TypeScript checks CI job does not install uv or Python, so the
module-level `execFileSync('uv', ...)` in schemas.contract.test.ts threw
ENOENT and failed the suite. Wrap the schema dump in a try/catch and
guard the describe block with `describe.skipIf` so the test skips in
environments without uv. Local dev and any CI job that has uv on PATH
still runs the cross-language contract assertion.
2026-05-15 02:11:04 +02:00
this . logger ? . error ( ` Error generating table description: ${ errorMessage ( error ) } ` , {
connectorId : input.connector.id ,
table : input.table.name ,
} ) ;
this . onWarning ? . ( {
code : 'enrichment_failed' ,
message : ` Failed to generate description for table ${ input . table . name } : ${ errorMessage ( error ) } ` ,
table : input.table.name ,
recoverable : true ,
metadata : { connectorId : input.connector.id } ,
} ) ;
return null ;
2026-05-10 23:12:26 +02:00
}
}
2026-05-12 14:34:15 +02:00
async generateDataSourceDescription ( input : KtxGenerateDataSourceDescriptionInput ) : Promise < string | null > {
2026-05-10 23:12:26 +02:00
if ( input . tables . length === 0 ) {
return 'No tables found in database' ;
}
const cacheKey = input . connectionName ? this . cache ? . buildConnectionKey ( input . connectionName ) : undefined ;
if ( cacheKey ) {
const cached = await this . cache ? . get ( cacheKey ) ;
if ( cached ) {
return cached ;
}
}
if ( ! input . connector . sampleTable ) {
2026-05-10 23:51:24 +02:00
this . logger ? . warn ( 'KTX scan connector does not support table sampling for data-source description generation' , {
2026-05-10 23:12:26 +02:00
connectorId : input.connector.id ,
} ) ;
return 'No accessible tables found in database' ;
}
const tablesToAnalyze = input . tables . slice ( 0 , 10 ) ;
const tableSamples = await runWithConcurrency ( tablesToAnalyze , this . settings . concurrencyLimit , async ( table ) = > {
try {
const sampleData = await input . connector . sampleTable ! (
{
connectionId : input.connectionId ,
table : toTableRef ( table ) ,
limit : 5 ,
} ,
input . context ,
) ;
2026-05-10 23:51:24 +02:00
return [ table . name , sampleData ] as [ string , KtxTableSampleResult ] ;
2026-05-10 23:12:26 +02:00
} catch ( error ) {
this . logger ? . warn ( ` Failed to sample table ' ${ table . name } ' for data source analysis - ${ errorMessage ( error ) } ` ) ;
return null ;
}
} ) ;
const accessibleSamples = tableSamples . filter (
2026-05-10 23:51:24 +02:00
( sample ) : sample is [ string , KtxTableSampleResult ] = > sample !== null ,
2026-05-10 23:12:26 +02:00
) ;
if ( accessibleSamples . length === 0 ) {
return 'No accessible tables found in database' ;
}
try {
2026-05-10 23:51:24 +02:00
const prompt = buildKtxDataSourceDescriptionPrompt ( {
2026-05-10 23:12:26 +02:00
tableSamples : accessibleSamples ,
dataSourceType : input.dataSourceType ,
2026-05-14 15:36:27 +02:00
maxWords : this.settings.dataSourceMaxWords ,
2026-05-10 23:12:26 +02:00
} ) ;
2026-05-14 15:36:27 +02:00
const description = await this . generateAiDescription ( prompt , 'ktx-data-source-description' ) ;
2026-05-12 14:34:15 +02:00
if ( cacheKey && description ) {
2026-05-10 23:12:26 +02:00
await this . cache ? . set ( cacheKey , description ) ;
}
return description ;
} catch ( error ) {
this . logger ? . error ( ` Error generating data source description: ${ errorMessage ( error ) } ` ) ;
return 'Failed to generate data source description' ;
}
}
private async generateOneColumnDescription (
2026-05-10 23:51:24 +02:00
input : KtxGenerateColumnDescriptionsInput ,
column : KtxDescriptionColumn ,
2026-05-10 23:12:26 +02:00
tableContext : string ,
) : Promise < ColumnTaskResult > {
const existingDescription = input . existingDescriptions ? . [ column . name ] ;
if ( input . skipExisting && existingDescription ) {
return {
columnName : column.name ,
description : existingDescription ,
skipped : true ,
processed : false ,
} ;
}
const tableRef = toTableRef ( input . table ) ;
const cacheKey = this . cache ? . buildColumnKey ( tableRef , column . name ) ;
if ( cacheKey ) {
const cached = await this . cache ? . get ( cacheKey ) ;
if ( cached ) {
return {
columnName : column.name ,
description : cached ,
skipped : true ,
processed : false ,
} ;
}
}
try {
let columnValues = column . sampleValues ;
if ( ! columnValues || columnValues . length === 0 ) {
if ( ! input . connector . sampleColumn ) {
fix(context): merge overlay columns onto manifest columns by name (#94)
* fix(context): merge overlay columns onto manifest columns by name
composeOverlay was appending overlay columns to the manifest column list,
producing duplicate entries when dbt/metabase overlays declared a column
just to attach descriptions. The duplicates carried no `type`, so the
pydantic SourceDefinition rejected them at semantic-query time and broke
`ktx sl query` for every overlay-backed measure. Now overlay columns
match base columns by name (case-insensitive): same-name entries merge
onto the manifest (overlay fields win, type/role fall back to the base,
descriptions merge per source key) and only new names append.
* refactor(sl): split overlay columns from column_overrides and enforce TS/Python wire contract
Overlay sources now have two distinct collections: `columns:` for computed
columns (requiring `expr` + `type`) and `column_overrides:` for metadata
patches to inherited manifest columns. Composing or loading an overlay that
mixes the two — or references an unknown column — fails with a typed error.
Introduce `ResolvedSemanticLayerSource` / `resolvedSourceSchema` /
`toResolvedWire` as the strict shape sent to the Python engine, and add a
schema contract test that diffs Zod against the Pydantic JSON schema dumped
by `python -m semantic_layer dump-schema`. `SourceDefinition` is now
`extra="forbid"` on the Python side.
`loadAllSources` surfaces per-file load errors instead of swallowing them,
so validation/query paths can report manifest shard parse failures.
* fix(context): make scan description generation resilient and quiet
A transient sampleTable failure during ingest used to take out every
table in a connection: generateTableDescription returned a hardcoded
'Table not found' string into descriptions.ai, and KtxDescriptionGenerator
was constructed without a logger, so the failure left no trail anywhere.
- sampleTable / sampleColumn calls retry 3x with 200/400/800ms backoff,
honouring KtxScanContext.signal via a new KtxAbortedError.
- On retry exhaustion or missing capability, table generation falls back
to a metadata-only prompt built from column name / native type / comment
/ rawDescriptions. The column path follows the same rule -- call the
LLM when any of samples or rawDescriptions are available; skip only
when both are absent.
- Logger is now threaded from KtxScanContext into the generator. Failures
emit structured KtxScanWarning entries (new description_fallback_used
code, plus existing sampling_failed / enrichment_failed /
connector_capability_missing). ktx scan groups warnings by code so a
batch of identical failures collapses to one summary line plus sample.
- Returns null on failure instead of the 'Table not found' sentinel; the
manifest writer's existing guard already skips empty descriptions, so
schema YAML no longer carries misleading text. SCAN_MANAGED_DESCRIPTION_KEYS
already strips stale 'ai' on merge, so existing YAML clears on next run.
Also suppress AI SDK v6 'system in messages' warning: pull system messages
out of KtxMessageBuilder.wrapSimple's output via a new splitKtxSystemMessages
helper and pass them top-level to generateText (preserves cacheControl
providerOptions on the SystemModelMessage). Agent-runner's local
splitSystemPromptMessages dedupes onto the shared helper.
* test(docs): align examples-docs assertions with revamped docs
PR #103 (setup/guide doc revamp) reworded several CLI examples and
connection labels; the assertions in scripts/examples-docs.test.mjs
still referenced the pre-revamp wording and were failing in CI on main.
Update the regexes to match the post-revamp content:
- drop the `--json` flag from the sl-query example expectation
- move the `Driver:` / `Status: ok` probe to the connection reference,
which is where that output now lives (driver id is lowercase
`postgres`, not the display name `PostgreSQL`)
- drop the obsolete `Install \`uv\`...` troubleshooting line
- accept `<connectionId>` everywhere; the docs no longer use the
hyphenated `<connection-id>` form
- match the `warehouse` connection id used in the quickstart instead of
the `postgres-warehouse` id only used in the README and setup ref
* fix(sl): skip TS/Python schema contract test when uv is unavailable
The TypeScript checks CI job does not install uv or Python, so the
module-level `execFileSync('uv', ...)` in schemas.contract.test.ts threw
ENOENT and failed the suite. Wrap the schema dump in a try/catch and
guard the describe block with `describe.skipIf` so the test skips in
environments without uv. Local dev and any CI job that has uv on PATH
still runs the cross-language contract assertion.
2026-05-15 02:11:04 +02:00
this . logger ? . warn ( 'KTX scan connector does not support column sampling; using available metadata only' , {
2026-05-10 23:12:26 +02:00
connectorId : input.connector.id ,
table : input.table.name ,
column : column.name ,
} ) ;
fix(context): merge overlay columns onto manifest columns by name (#94)
* fix(context): merge overlay columns onto manifest columns by name
composeOverlay was appending overlay columns to the manifest column list,
producing duplicate entries when dbt/metabase overlays declared a column
just to attach descriptions. The duplicates carried no `type`, so the
pydantic SourceDefinition rejected them at semantic-query time and broke
`ktx sl query` for every overlay-backed measure. Now overlay columns
match base columns by name (case-insensitive): same-name entries merge
onto the manifest (overlay fields win, type/role fall back to the base,
descriptions merge per source key) and only new names append.
* refactor(sl): split overlay columns from column_overrides and enforce TS/Python wire contract
Overlay sources now have two distinct collections: `columns:` for computed
columns (requiring `expr` + `type`) and `column_overrides:` for metadata
patches to inherited manifest columns. Composing or loading an overlay that
mixes the two — or references an unknown column — fails with a typed error.
Introduce `ResolvedSemanticLayerSource` / `resolvedSourceSchema` /
`toResolvedWire` as the strict shape sent to the Python engine, and add a
schema contract test that diffs Zod against the Pydantic JSON schema dumped
by `python -m semantic_layer dump-schema`. `SourceDefinition` is now
`extra="forbid"` on the Python side.
`loadAllSources` surfaces per-file load errors instead of swallowing them,
so validation/query paths can report manifest shard parse failures.
* fix(context): make scan description generation resilient and quiet
A transient sampleTable failure during ingest used to take out every
table in a connection: generateTableDescription returned a hardcoded
'Table not found' string into descriptions.ai, and KtxDescriptionGenerator
was constructed without a logger, so the failure left no trail anywhere.
- sampleTable / sampleColumn calls retry 3x with 200/400/800ms backoff,
honouring KtxScanContext.signal via a new KtxAbortedError.
- On retry exhaustion or missing capability, table generation falls back
to a metadata-only prompt built from column name / native type / comment
/ rawDescriptions. The column path follows the same rule -- call the
LLM when any of samples or rawDescriptions are available; skip only
when both are absent.
- Logger is now threaded from KtxScanContext into the generator. Failures
emit structured KtxScanWarning entries (new description_fallback_used
code, plus existing sampling_failed / enrichment_failed /
connector_capability_missing). ktx scan groups warnings by code so a
batch of identical failures collapses to one summary line plus sample.
- Returns null on failure instead of the 'Table not found' sentinel; the
manifest writer's existing guard already skips empty descriptions, so
schema YAML no longer carries misleading text. SCAN_MANAGED_DESCRIPTION_KEYS
already strips stale 'ai' on merge, so existing YAML clears on next run.
Also suppress AI SDK v6 'system in messages' warning: pull system messages
out of KtxMessageBuilder.wrapSimple's output via a new splitKtxSystemMessages
helper and pass them top-level to generateText (preserves cacheControl
providerOptions on the SystemModelMessage). Agent-runner's local
splitSystemPromptMessages dedupes onto the shared helper.
* test(docs): align examples-docs assertions with revamped docs
PR #103 (setup/guide doc revamp) reworded several CLI examples and
connection labels; the assertions in scripts/examples-docs.test.mjs
still referenced the pre-revamp wording and were failing in CI on main.
Update the regexes to match the post-revamp content:
- drop the `--json` flag from the sl-query example expectation
- move the `Driver:` / `Status: ok` probe to the connection reference,
which is where that output now lives (driver id is lowercase
`postgres`, not the display name `PostgreSQL`)
- drop the obsolete `Install \`uv\`...` troubleshooting line
- accept `<connectionId>` everywhere; the docs no longer use the
hyphenated `<connection-id>` form
- match the `warehouse` connection id used in the quickstart instead of
the `postgres-warehouse` id only used in the README and setup ref
* fix(sl): skip TS/Python schema contract test when uv is unavailable
The TypeScript checks CI job does not install uv or Python, so the
module-level `execFileSync('uv', ...)` in schemas.contract.test.ts threw
ENOENT and failed the suite. Wrap the schema dump in a try/catch and
guard the describe block with `describe.skipIf` so the test skips in
environments without uv. Local dev and any CI job that has uv on PATH
still runs the cross-language contract assertion.
2026-05-15 02:11:04 +02:00
columnValues = [ ] ;
} else {
const sampleColumn = input . connector . sampleColumn ;
try {
const sample = await retryAsync (
( ) = >
sampleColumn (
{
connectionId : input.connectionId ,
table : tableRef ,
column : column.name ,
limit : 50 ,
} ,
input . context ,
) ,
{
attempts : 3 ,
baseDelayMs : 200 ,
signal : input.context.signal ,
onAttemptFailure : ( error , attempt ) = > {
this . logger ? . warn (
` sampleColumn attempt ${ attempt } failed for ${ input . table . name } . ${ column . name } : ${ errorMessage ( error ) } ` ,
{
connectorId : input.connector.id ,
table : input.table.name ,
column : column.name ,
attempt ,
} ,
) ;
} ,
} ,
) ;
columnValues = sample . values ;
} catch ( error ) {
if ( error instanceof KtxAbortedError ) {
throw error ;
}
this . logger ? . warn (
` sampleColumn exhausted retries for ${ input . table . name } . ${ column . name } ; using available metadata only: ${ errorMessage ( error ) } ` ,
{
connectorId : input.connector.id ,
table : input.table.name ,
column : column.name ,
} ,
) ;
columnValues = [ ] ;
}
2026-05-10 23:12:26 +02:00
}
}
const nonNullValues = ( columnValues ? ? [ ] ) . filter ( ( value ) = > value !== null && value !== undefined ) ;
fix(context): merge overlay columns onto manifest columns by name (#94)
* fix(context): merge overlay columns onto manifest columns by name
composeOverlay was appending overlay columns to the manifest column list,
producing duplicate entries when dbt/metabase overlays declared a column
just to attach descriptions. The duplicates carried no `type`, so the
pydantic SourceDefinition rejected them at semantic-query time and broke
`ktx sl query` for every overlay-backed measure. Now overlay columns
match base columns by name (case-insensitive): same-name entries merge
onto the manifest (overlay fields win, type/role fall back to the base,
descriptions merge per source key) and only new names append.
* refactor(sl): split overlay columns from column_overrides and enforce TS/Python wire contract
Overlay sources now have two distinct collections: `columns:` for computed
columns (requiring `expr` + `type`) and `column_overrides:` for metadata
patches to inherited manifest columns. Composing or loading an overlay that
mixes the two — or references an unknown column — fails with a typed error.
Introduce `ResolvedSemanticLayerSource` / `resolvedSourceSchema` /
`toResolvedWire` as the strict shape sent to the Python engine, and add a
schema contract test that diffs Zod against the Pydantic JSON schema dumped
by `python -m semantic_layer dump-schema`. `SourceDefinition` is now
`extra="forbid"` on the Python side.
`loadAllSources` surfaces per-file load errors instead of swallowing them,
so validation/query paths can report manifest shard parse failures.
* fix(context): make scan description generation resilient and quiet
A transient sampleTable failure during ingest used to take out every
table in a connection: generateTableDescription returned a hardcoded
'Table not found' string into descriptions.ai, and KtxDescriptionGenerator
was constructed without a logger, so the failure left no trail anywhere.
- sampleTable / sampleColumn calls retry 3x with 200/400/800ms backoff,
honouring KtxScanContext.signal via a new KtxAbortedError.
- On retry exhaustion or missing capability, table generation falls back
to a metadata-only prompt built from column name / native type / comment
/ rawDescriptions. The column path follows the same rule -- call the
LLM when any of samples or rawDescriptions are available; skip only
when both are absent.
- Logger is now threaded from KtxScanContext into the generator. Failures
emit structured KtxScanWarning entries (new description_fallback_used
code, plus existing sampling_failed / enrichment_failed /
connector_capability_missing). ktx scan groups warnings by code so a
batch of identical failures collapses to one summary line plus sample.
- Returns null on failure instead of the 'Table not found' sentinel; the
manifest writer's existing guard already skips empty descriptions, so
schema YAML no longer carries misleading text. SCAN_MANAGED_DESCRIPTION_KEYS
already strips stale 'ai' on merge, so existing YAML clears on next run.
Also suppress AI SDK v6 'system in messages' warning: pull system messages
out of KtxMessageBuilder.wrapSimple's output via a new splitKtxSystemMessages
helper and pass them top-level to generateText (preserves cacheControl
providerOptions on the SystemModelMessage). Agent-runner's local
splitSystemPromptMessages dedupes onto the shared helper.
* test(docs): align examples-docs assertions with revamped docs
PR #103 (setup/guide doc revamp) reworded several CLI examples and
connection labels; the assertions in scripts/examples-docs.test.mjs
still referenced the pre-revamp wording and were failing in CI on main.
Update the regexes to match the post-revamp content:
- drop the `--json` flag from the sl-query example expectation
- move the `Driver:` / `Status: ok` probe to the connection reference,
which is where that output now lives (driver id is lowercase
`postgres`, not the display name `PostgreSQL`)
- drop the obsolete `Install \`uv\`...` troubleshooting line
- accept `<connectionId>` everywhere; the docs no longer use the
hyphenated `<connection-id>` form
- match the `warehouse` connection id used in the quickstart instead of
the `postgres-warehouse` id only used in the README and setup ref
* fix(sl): skip TS/Python schema contract test when uv is unavailable
The TypeScript checks CI job does not install uv or Python, so the
module-level `execFileSync('uv', ...)` in schemas.contract.test.ts threw
ENOENT and failed the suite. Wrap the schema dump in a try/catch and
guard the describe block with `describe.skipIf` so the test skips in
environments without uv. Local dev and any CI job that has uv on PATH
still runs the cross-language contract assertion.
2026-05-15 02:11:04 +02:00
const hasRawDescriptions = descriptionSources ( column . rawDescriptions ) . length > 0 ;
if ( nonNullValues . length === 0 && ! hasRawDescriptions ) {
2026-05-10 23:12:26 +02:00
return {
columnName : column.name ,
description : null ,
skipped : false ,
processed : false ,
} ;
}
2026-05-10 23:51:24 +02:00
const prompt = buildKtxColumnDescriptionPrompt ( {
2026-05-10 23:12:26 +02:00
columnName : column.name ,
columnValues : nonNullValues ,
tableContext ,
dataSourceType : input.dataSourceType ,
supportsNestedAnalysis : input.supportsNestedAnalysis ,
rawDescriptions : column.rawDescriptions ,
2026-05-14 15:36:27 +02:00
maxWords : this.settings.columnMaxWords ,
2026-05-10 23:12:26 +02:00
} ) ;
2026-05-14 15:36:27 +02:00
const description = await this . generateAiDescription ( prompt , 'ktx-column-description' ) ;
2026-05-10 23:12:26 +02:00
2026-05-12 14:34:15 +02:00
if ( cacheKey && description ) {
2026-05-10 23:12:26 +02:00
await this . cache ? . set ( cacheKey , description ) ;
}
return {
columnName : column.name ,
description ,
skipped : false ,
2026-05-12 14:34:15 +02:00
processed : description !== null ,
2026-05-10 23:12:26 +02:00
} ;
} catch ( error ) {
fix(context): merge overlay columns onto manifest columns by name (#94)
* fix(context): merge overlay columns onto manifest columns by name
composeOverlay was appending overlay columns to the manifest column list,
producing duplicate entries when dbt/metabase overlays declared a column
just to attach descriptions. The duplicates carried no `type`, so the
pydantic SourceDefinition rejected them at semantic-query time and broke
`ktx sl query` for every overlay-backed measure. Now overlay columns
match base columns by name (case-insensitive): same-name entries merge
onto the manifest (overlay fields win, type/role fall back to the base,
descriptions merge per source key) and only new names append.
* refactor(sl): split overlay columns from column_overrides and enforce TS/Python wire contract
Overlay sources now have two distinct collections: `columns:` for computed
columns (requiring `expr` + `type`) and `column_overrides:` for metadata
patches to inherited manifest columns. Composing or loading an overlay that
mixes the two — or references an unknown column — fails with a typed error.
Introduce `ResolvedSemanticLayerSource` / `resolvedSourceSchema` /
`toResolvedWire` as the strict shape sent to the Python engine, and add a
schema contract test that diffs Zod against the Pydantic JSON schema dumped
by `python -m semantic_layer dump-schema`. `SourceDefinition` is now
`extra="forbid"` on the Python side.
`loadAllSources` surfaces per-file load errors instead of swallowing them,
so validation/query paths can report manifest shard parse failures.
* fix(context): make scan description generation resilient and quiet
A transient sampleTable failure during ingest used to take out every
table in a connection: generateTableDescription returned a hardcoded
'Table not found' string into descriptions.ai, and KtxDescriptionGenerator
was constructed without a logger, so the failure left no trail anywhere.
- sampleTable / sampleColumn calls retry 3x with 200/400/800ms backoff,
honouring KtxScanContext.signal via a new KtxAbortedError.
- On retry exhaustion or missing capability, table generation falls back
to a metadata-only prompt built from column name / native type / comment
/ rawDescriptions. The column path follows the same rule -- call the
LLM when any of samples or rawDescriptions are available; skip only
when both are absent.
- Logger is now threaded from KtxScanContext into the generator. Failures
emit structured KtxScanWarning entries (new description_fallback_used
code, plus existing sampling_failed / enrichment_failed /
connector_capability_missing). ktx scan groups warnings by code so a
batch of identical failures collapses to one summary line plus sample.
- Returns null on failure instead of the 'Table not found' sentinel; the
manifest writer's existing guard already skips empty descriptions, so
schema YAML no longer carries misleading text. SCAN_MANAGED_DESCRIPTION_KEYS
already strips stale 'ai' on merge, so existing YAML clears on next run.
Also suppress AI SDK v6 'system in messages' warning: pull system messages
out of KtxMessageBuilder.wrapSimple's output via a new splitKtxSystemMessages
helper and pass them top-level to generateText (preserves cacheControl
providerOptions on the SystemModelMessage). Agent-runner's local
splitSystemPromptMessages dedupes onto the shared helper.
* test(docs): align examples-docs assertions with revamped docs
PR #103 (setup/guide doc revamp) reworded several CLI examples and
connection labels; the assertions in scripts/examples-docs.test.mjs
still referenced the pre-revamp wording and were failing in CI on main.
Update the regexes to match the post-revamp content:
- drop the `--json` flag from the sl-query example expectation
- move the `Driver:` / `Status: ok` probe to the connection reference,
which is where that output now lives (driver id is lowercase
`postgres`, not the display name `PostgreSQL`)
- drop the obsolete `Install \`uv\`...` troubleshooting line
- accept `<connectionId>` everywhere; the docs no longer use the
hyphenated `<connection-id>` form
- match the `warehouse` connection id used in the quickstart instead of
the `postgres-warehouse` id only used in the README and setup ref
* fix(sl): skip TS/Python schema contract test when uv is unavailable
The TypeScript checks CI job does not install uv or Python, so the
module-level `execFileSync('uv', ...)` in schemas.contract.test.ts threw
ENOENT and failed the suite. Wrap the schema dump in a try/catch and
guard the describe block with `describe.skipIf` so the test skips in
environments without uv. Local dev and any CI job that has uv on PATH
still runs the cross-language contract assertion.
2026-05-15 02:11:04 +02:00
if ( error instanceof KtxAbortedError ) {
throw error ;
}
this . logger ? . error ( ` Error analyzing column ' ${ column . name } ': ${ errorMessage ( error ) } ` , {
connectorId : input.connector.id ,
table : input.table.name ,
column : column.name ,
} ) ;
2026-05-10 23:12:26 +02:00
return {
columnName : column.name ,
2026-05-12 14:34:15 +02:00
description : null ,
2026-05-10 23:12:26 +02:00
skipped : false ,
processed : false ,
} ;
}
}
2026-05-14 15:36:27 +02:00
private async generateAiDescription ( prompt : KtxDescriptionPrompt , _operationName : string ) : Promise < string | null > {
2026-05-10 23:12:26 +02:00
try {
2026-05-10 23:51:24 +02:00
const text = await generateKtxText ( {
2026-05-10 23:12:26 +02:00
llmProvider : this.llmProvider ,
role : 'candidateExtraction' ,
2026-05-14 15:36:27 +02:00
system : prompt.system ,
prompt : prompt.user ,
2026-05-10 23:12:26 +02:00
temperature : this.settings.temperature ,
} ) ;
const description = text . trim ( ) ;
2026-05-12 14:34:15 +02:00
return description || null ;
2026-05-10 23:12:26 +02:00
} catch ( error ) {
this . logger ? . error ( ` Error generating AI description: ${ errorMessage ( error ) } ` ) ;
2026-05-12 14:34:15 +02:00
return null ;
2026-05-10 23:12:26 +02:00
}
}
}