SurfSense/automation-design-plan.md

57 KiB
Raw Blame History

SurfSense Automation Feature — Design Plan (v2)

A generic, extensible automation system for SurfSense that lets users (and future SurfSense features) trigger agent work on a schedule, on an external event, or on demand — with the ability to author automations either by hand or from a natural-language description that yields an editable, structured definition.

This document supersedes the v1 draft. It folds in the design audit pass and the corrections from working through worked examples (notably: removing the connector bias, clarifying the executor's role, integrating MCP cleanly, and committing to JSON Schema as the single declarative language).


1. The load-bearing principle

The JSON definition is the program. Everything else is interpreter.

Every decision in this document serves that principle. If we ever face a design choice and one option lets some behavior leak out of the definition into the engine, we pick the other option.

Three properties follow from this principle, and they're the reason the system will survive feature growth:

  • Reproducibility — same definition + same inputs → same observable behavior, regardless of which version of the engine runs it.
  • Portability — definitions can be exported, imported, version- controlled, code-reviewed, and shared across SurfSense instances.
  • LLM tractability — the NL authoring flow works because the LLM only needs to produce a self-contained JSON document that validates against a schema. It doesn't need to understand the engine.

2. The three-layer contract

The system is structured as three layers. Layers 1 and 3 are defined by SurfSense developers (at registration time). Layer 2 is what users write (or the NL generator produces). The runtime reads all three to do its job.

Layer What it is Defined by
1. Action contract Per-action params and output schema Developers, at startup
2. Automation definition One concrete saved automation Users (or NL generator)
3. Trigger contract Per-trigger params and payload schemas Developers, at startup

Each layer constrains the next. The runtime reads all three but doesn't know what's in them ahead of time. That's how a new action or trigger type becomes available across the engine without code changes outside its registration.

A unification layer below Layer 1 — one catalog of "things this SurfSense instance can do," shared by automations, agents, and future surfaces — was considered and deferred (§3). v1 actions are stand-alone.

Schema language

Every shape in every layer is described in JSON Schema (draft 2020-12). No exceptions, no parallel languages, no inline shorthand. Two documented extensions on top:

  • default: "$some_token" — runtime-resolved defaults. The vocabulary is fixed: $last_fired_at, $creator, $space_default. The engine resolves these to values before validation.
  • x-surfsense-* annotations — editor hints (widget type, autocomplete source). The validator ignores them; the form editor reads them.

3. Capability unification layer — deferred to post-v1

Earlier drafts introduced a Capability registry as Layer 1: one catalog of "things this SurfSense instance can do," shared by the automation engine (as actions), the agent (as tools), and any future HTTP surface. The motivation is real — one source of truth beats N parallel registries — but v1 has a single action (agent_task) and a single consumer (the automation engine). The five-field shape sketched earlier (id, description, input_schema, output_schema, handler) cannot safely host any non-trivial capability: it carries no caller identity, no search-space scoping, and no authorization gate on tool delegation. Building the abstraction with one consumer would lock in a shape that doesn't survive the second consumer.

The unification layer returns when the second consumer lands (Phase 2 tight actions or Phase 4 MCP), redesigned from the start with:

  • A CallContext carrying caller user id, search space id, and run id, passed to every handler invocation.
  • Explicit scope declarations per capability (e.g. reads:documents, writes:slack, destructive) for the authorization layer to read.
  • A per-user, per-search-space filter consulted at both definition save time (validating agent_task.tools) and run time (scoping the agent's tool list to what the automation creator can delegate).

Until then:

  • v1 actions are stand-alone units (Layer 1 below); the automation engine reads its own action registry, nothing else.
  • agent_task.params.tools is a forward-looking allowlist field with no v1 semantics beyond "list of string identifiers." The handler's tool resolution is opaque to the automation contract.

Credentials — deferred to Phase 2

External-credential handlers (Slack, email, etc.) require per-user or per-connection auth. v1 actions run server-side with app-level configuration. When tight actions ship in Phase 2, the credential design lands as part of the unification redesign: connection IDs in the definition (never tokens); credentials loaded per-call by the handler context (never pre-loaded into worker memory); credentials never enter LLM context.

MCP — deferred to Phase 4

External tool servers feeding tools into a shared registry land with the rest of the integration tooling in Phase 4, after the unification layer is in place. The two-tier registry, mcp_connections and mcp_tools tables, and the harvester arrive as a single coherent step then.


4. Action contract

An Action is what a user references in a plan step. Some actions are deterministic single-purpose handlers (slack_post, send_email); one action (agent_task) hosts an LLM and a tool allowlist for cases where judgment is needed. The contract is the same in both cases — only the handler differs.

@dataclass(frozen=True, slots=True)
class ActionDefinition:
    type: str            # "agent_task", "slack_post"
    name: str            # short UI label
    description: str     # for the NL generator and the UI
    params_schema: dict  # JSON Schema for step.params
    handler: ActionHandler

This is the v1 shape: five fields, no handler context, no output contract, no artifact declaration. The deferrals are intentional:

  • output_contract — Phase 2. Deterministic handlers will return a fixed shape; v1's only action (agent_task) takes an output_schema inside params and validates against that instead.
  • produces_artifacts — Phase 5. Artifact lifecycle (storage, signed URLs, retention) is its own design step; v1 handlers persist their own outputs.
  • Handler context — paired with the unification redesign (§3). v1 handlers receive (args) only; per-user / per-search-space behavior is not yet a v1 concern.

Tight vs loose actions

Two patterns coexist by design:

  • Tight actions (slack_post, linear_create_issue, send_email) — deterministic single-purpose handlers. ~20 LOC each. Phase 2.
  • Loose actions (agent_task) — params_schema accepts a prompt, a tools allowlist, and an optional output_schema declaring what the agent must return; the handler validates the agent's output against it. v1.

The agent's tools allowlist resolves opaquely in v1; the redesigned unification layer (§3) will give both invocation modes access to the same vocabulary, with per-user authorization gating both.

How names in the definition become function calls

The definition contains strings like "action": "agent_task". The string is just a name — it does not point to a function. At runtime, the executor performs a name-based lookup against the action registry:

action_def = action_registry.get(step.action)   # dict lookup
handler = action_def.handler                    # Python callable
result = await handler(resolved_params)         # invocation

The registry is a Python dict populated at process startup. Each entry in automations/registries/actions/*.py calls register_action(...) at module import time, putting its ActionDefinition (including the handler function reference) into the registry.

The definition is pure data. The registry is the engine's runtime vocabulary. They meet at name-based lookup; nothing else crosses the boundary.

The full expressive spectrum

The contract supports a continuous spectrum from purely deterministic to fully agentic. Six practical shapes worth recognizing:

Shape Example Cost / latency profile
1. Direct call slack_post with literal channel and template No LLM. ~200ms. Fractions of a cent.
2. Direct call with computed inputs linear_create_issue using {{summary.title}} from a prior step No LLM for this step. Cheap.
3. Single-domain agent task agent_task with tools: ["slack.*"] only One LLM, bounded toolset.
4. Multi-domain agent task, narrow agent_task with tools: ["github.list_pull_requests", "linear.create_issue"] One LLM, named tools.
5. Multi-domain agent task, broad agent_task with tools: ["slack.*", "github.*", "linear.*"] One LLM, large toolset, most agentic.
6. Composed plan agent_task (narrow) for thinking → slack_post + linear_create_issue for acting Best cost-to-power ratio.

Shape 6 is the underrated one and the cost-and-speed answer. The agent reasons once (Shape 3 or 4) and its structured output drives several deterministic actions. This is roughly 510x cheaper and 34x faster than forcing the agent to do everything (Shape 5) and produces the same outcome.

The NL generator's job is to propose Shape 6-style plans by default. The Review LLM flags proposals that use agent_task for steps a deterministic action could handle. This is the discipline that keeps automations cheap at scale.

The user navigates the spectrum by intent (describing what they want), not by mechanism — the shape selection is the engine's responsibility, not the user's.


5. Automation definition

This is the JSON the user writes (or the NL generator produces). Stored in automations.definition as JSONB.

Top-level shape

{
  "schema_version": "1.0",
  "name": "Daily competitor digest",
  "goal": "Summarize new competitor content and post to Slack",

  "inputs": {
    "schema": {
      "type": "object",
      "required": ["since"],
      "properties": {
        "since": { "type": "string", "format": "date-time",
                   "default": "$last_fired_at" },
        "tags":  { "type": "array", "items": { "type": "string" },
                   "default": ["competitor"] }
      }
    }
  },

  "triggers": [
    {
      "type": "schedule",
      "params": { "cron": "0 9 * * 1-5", "timezone": "Africa/Kigali" }
    }
  ],

  "plan": [
    {
      "step_id": "research",
      "action": "agent_task",
      "params": {
        "prompt": "Find documents tagged {{inputs.tags}} indexed since {{inputs.since}}. Return JSON with bullets and source_doc_ids.",
        "tools": ["search_space.query", "search_space.fetch_document"],
        "model": "anthropic/claude-sonnet-4-7",
        "output_schema": {
          "type": "object",
          "required": ["bullets", "source_doc_ids"],
          "properties": {
            "bullets":        { "type": "array", "items": { "type": "string" } },
            "source_doc_ids": { "type": "array", "items": { "type": "string" } }
          }
        }
      },
      "output_as": "summary"
    },
    {
      "step_id": "deliver",
      "action": "slack_post",
      "params": {
        "channel_id": "C0123",
        "message_template": "*Competitor digest*\n\n{% for b in summary.bullets %}• {{b}}\n{% endfor %}"
      }
    }
  ],

  "execution": {
    "timeout_seconds": 600,
    "max_retries": 2,
    "retry_backoff": "exponential",
    "concurrency": "drop_if_running",
    "on_failure": [ /* steps to run if main plan fails after retries */ ]
  },

  "metadata": { "tags": ["digest"] }
}

Plan steps

{
  "step_id": "...",                      // unique within plan
  "action": "...",                       // references an ActionDefinition.type
  "when": "{{ ... }}",                   // optional Jinja expr → bool; false = skip
  "params": { ... },                     // validated against action's params_schema
  "output_as": "...",                    // binds output to this name for later steps
  "max_retries": 0,                      // optional, overrides automation default
  "timeout_seconds": 1200                // optional, overrides automation default
}

Steps run sequentially. No parallelism, no DAGs, no loops. If a user needs branching, they use when: on multiple steps. If they need parallelism or iteration, they use agent_task and let the agent reason about it, or they compose automations through events (§7.5).


6. Trigger contract

Three trigger types. That's the entire taxonomy.

schedule

TriggerDefinition(
    type="schedule",
    params_model=ScheduleTriggerParams,  # cron + timezone
)
# At fire time the schedule producer emits runtime inputs
# (fired_at, scheduled_for, last_fired_at) which are merged with the
# trigger row's static_inputs (static wins) and validated against
# automation.definition.inputs.schema_.

Implementation: extends app/utils/periodic_scheduler.py, which already reads connector sync schedules. Adds a second source — automation_triggers WHERE type='schedule'. Same Celery Beat checker, two source tables.

Minimum interval: 1 minute (the existing checker's resolution). The form editor warns when users set intervals under 15 minutes that they probably want an event trigger instead.

webhook

TriggerDefinition(
    type="webhook",
    params_schema={
        "type": "object",
        "properties": {
            "input_mapping": {
                "type": "object",
                "additionalProperties": { "type": "string" }
                # values are JSONPath expressions
            }
        }
    },
    # payload is whatever the POST body is; user-defined shape via mapping
)

Endpoint: POST /api/v1/automations/{id}/fire. Bearer token shown once, hashed at rest, rotatable, revocable. Returns 202 Accepted with the created run's URL. Caller polls for status; we do not push callbacks in v1 (a callback_webhook action can be added later).

Idempotency: honors Idempotency-Key header or idempotency_key in body. Dedups against runs in the last 24 hours.

event

TriggerDefinition(
    type="event",
    params_schema={
        "type": "object",
        "required": ["event_type"],
        "properties": {
            "event_type": { "type": "string" },   # e.g. "drive.file_added"
                                                   # or "surfsense.podcast.generated"
            "filters":    { "$ref": "#/definitions/filter_expression" }
        }
    }
    # payload shape is documented per event_type in a separate registry
)

Events absorb both connector events and internal SurfSense events. A file added to Drive and a podcast finishing in SurfSense are both events in the same domain_events table, both subscribable by automations, both matched by the same dispatcher code. The engine doesn't distinguish.

Filter grammar

Filters are JSON-structured operators, not expressions. This is the one place we deliberately don't use Jinja, because filters run on a hot path (every event matched against every subscribing trigger) and structured filters can be indexed and short-circuited.

Vocabulary:

  • Equality: equals, not_equals
  • String: starts_with, ends_with, contains, regex
  • Numeric: gt, gte, lt, lte
  • Set: in, not_in
  • Existence: exists
  • Composition: $and, $or, $not

Inspired by AWS EventBridge and MongoDB query syntax. The filter grammar itself is published as a JSON Schema, so users get inline error messages.


7. Runtime components

Each component is distinct, replaceable, and has one job.

7.1 Dispatcher

What it does: matches firing triggers to automations, creates AutomationRun rows, enqueues executor tasks.

For schedule triggers: Celery Beat polls the trigger table, computes due ones, fires.

For webhook triggers: the FastAPI handler is the dispatcher entry point. Validates token, runs input_mapping, creates run.

For event triggers: subscribes to the domain_events table. For each new event, evaluates all matching triggers' filters, fires the matches.

Common path (after a trigger has fired):

  1. Resolve inputs from trigger payload and defaults
  2. Validate resolved inputs against the automation's input schema
  3. Idempotency check — dedup against existing pending/running runs
  4. Snapshot the resolved definition into the run row (immutable history)
  5. Enqueue executor task on the single automations_default Celery queue

The cost-estimate pre-check (originally step 3) is deferred. v1 actions do not declare cost estimates, the run row has no cost_usd column, and no handler reports tokens used — so neither pre-flight prediction nor mid-flight accumulation can be enforced. Execution therefore does not expose budget_cap_usd in v1; it returns as a single field addition the day the cost ledger ships (per-action cost reporting

  • automation_runs.cost_usd column + executor accumulation).

Queue routing by expected_duration_seconds is deferred until load patterns justify a second queue. v1 uses a single queue.

7.2 Executor

What it is: a Celery task wrapping a single function that walks a plan step by step. Not an agent, not a workflow engine, not a scheduler. A loop with bookkeeping. Maybe 200 lines.

async def execute_run(run_id: int) -> None:
    run = load_run(run_id); run.status = "running"; save(run)
    context = build_run_context(run)
    step_outputs = {}

    for step in run.plan:
        if step.when and not evaluate_predicate(step.when, context | step_outputs):
            record_step_skipped(run, step); continue

        resolved_params = render_params(step.params, context | step_outputs)
        action = action_registry.get(step.action)
        validate(resolved_params, action.params_schema)

        try:
            result = await with_retries(
                action.handler,
                ctx=build_action_context(run, action),
                args=resolved_params,
                policy=step.retry_policy or run.execution.retry_policy,
            )
            validate(result, step.output_schema)
            if step.output_as:
                step_outputs[step.output_as] = result
            record_step_succeeded(run, step, result)
        except Exception as e:
            record_step_failed(run, step, e)
            await run_on_failure(run, e)
            return

    run.status = "succeeded"; save(run)
    publish_event("automation.run.succeeded", run)   # see §7.5

Intelligence lives inside handlers, not in the executor. The most intelligent handler is agent_task, which spins up a LangGraph Deep Agent for one step and returns when the agent finishes. The executor sees a validated dict come back; it doesn't know that step was "smart."

7.3 Action handlers

One handler per ActionDefinition.type. Receives the validated args dict and returns whatever the step's output validates against (a fixed shape declared by tight actions, or a dynamic shape declared via output_schema in the step params for agent_task).

Handlers do not know about retries or timeouts — those are the executor's concern.

In v1, handlers take (args) only. The CallContext parameter sketched in §7.2's pseudo-code (caller user id, search space id, run id, credential resolver) arrives with the unification layer redesign (§3); v1's single action (agent_task) reads what it needs from app-level configuration.

7.4 Template engine

Why it exists

Most fields in an automation definition contain literal strings the user authored once — but the actual rendered value has to change per run, because it includes data from the trigger payload or from prior step outputs. The template engine is what turns "Daily digest for {{run.started_at}}" into "Daily digest for 2026-05-26" at run time.

Three fields use it:

  • *_template strings in tight action configs (Slack messages, email bodies, Linear titles, etc.)
  • prompt in agent_task configs (so the agent sees resolved values, not {{...}} placeholders)
  • when: step predicates (which need to evaluate to a boolean)

Public interface

Single module, ~80 lines. Three public functions — everything else in the engine routes through these:

def render_template(template: str, context: dict) -> str: ...
def evaluate_predicate(expression: str, context: dict) -> bool: ...
def build_run_context(run, step_outputs) -> dict: ...

Backed by Jinja2's SandboxedEnvironment. The whole module is the seam: if the template language is ever swapped, only this file changes.

Security architecture: allowlist by default

SandboxedEnvironment starts empty. A freshly-created instance gives a template access to:

  • Variables in the context dict we pass in (run, inputs, prior step outputs)
  • Public (non-underscore) attributes of those variables
  • Jinja's built-in control flow ({% if %}, {% for %}, {% set %})

Nothing else. No Python builtins, no modules, no I/O, no network, no filesystem. Everything beyond the above must be explicitly registered. This is the structurally important property: anything we didn't add is inaccessible. The risk surface equals the size of what we registered.

The three sandbox rules that enforce this:

  1. Attribute access is filtered — names starting with underscore are rejected. This blocks the entire family of {{x.__class__.__mro__...}} Python escape paths in one rule.
  2. Globals are allowlist-onlyopen, eval, exec, __import__, getattr, every module name, are all absent unless we register them. We register zero globals.
  3. Unsafe callables are blockedstr.format and str.format_map specifically (due to CVE-2016-10745), plus anything marked unsafe_callable.

What we register, exactly

  • Filters: a curated 15, no more. join, length, default, upper, lower, truncate, tojson, date, replace, trim, slugify, first, last, sort, reverse. Each one is audited for what it does with its input; none of them takes a callable, runs eval, or reaches into Python objects beyond simple data transformation.
  • Globals: none.
  • Tests: only the safe built-ins (defined, none, number, string, mapping, sequence, boolean).

Adding a new filter requires a deliberate code change and review: does this filter do anything dangerous with its input? If yes, don't add it. The list only grows by audited additions.

Runtime limits (defense in depth)

The sandbox handles the attack surface inside the template language. Three additional limits handle resource exhaustion that the language permits but the runtime shouldn't tolerate:

  • Template source length capped at 8 KB. Checked before parsing.
  • Render time capped at 100 ms per render. Implemented via a watchdog thread; renders that exceed are killed and the step fails. Catches {% for i in range(10**9) %} and nested loop bombs.
  • Output size capped at 1 MB. A small template can produce a multi-GB string via {{ 'A' * 10**8 }}-style multiplication; this catches it.

Plus StrictUndefined: any reference to a missing variable raises immediately rather than silently rendering empty, so misconfigurations fail fast.

Threat model and residual risk

The trust model from day one is:

  • Templates are generated by an LLM from a user's natural-language input (see §10), or written/edited by humans in the editable form
  • A second LLM reviews the proposal and produces a plain-language summary plus flagged anomalies for the user
  • The user reviews and approves before the automation runs
  • The Generator LLM's input is scoped (user prompt + schema + registry only — no arbitrary document content), minimizing prompt-injection paths

The sandbox + runtime limits + curated filter list protect against the malformed-template attack. Human review protects against the semantically-malicious-but-syntactically-valid attack. These are complementary layers, not redundant.

Known residual risks, each genuinely small:

  • Future Jinja CVEs. Historical sandbox bypasses have existed and been patched. This is a generic third-party-dependency risk, comparable to bugs in any other library we rely on. Mitigation: subscribe to security advisories, ship updates within a week of disclosure.
  • Side channels via prompts to LLMs. A template that renders into a prompt can attempt prompt injection of the agent at run time. This is not a sandbox concern but a separate concern in agent_task's design.
  • Operator deployments with long-lived secrets in worker env vars. Mitigation: credentials fetched per-handler-per-call via ActionContext.resolve_credentials, never pre-loaded into worker env vars accessible to templates.

The sandbox-with-allowlist architecture means the attack surface equals the set of things we registered. With zero globals registered and 15 audited filters, the surface is small, bounded, and reviewable. This is the structural property that makes the architecture sound, and it doesn't depend on hypothetical assumptions about who authors templates.

Pre-Phase-5 gate

One trust-model change is documented in the roadmap: Phase 5 introduces template sharing across SearchSpaces (automation templates as exportable, importable artifacts). At that point, the approver of a template (the original author) is no longer the runner (the importer). The "human reviews before save" mitigation breaks down because the reviewer doesn't bear the risk.

Before Phase 5 ships, this needs an explicit re-approval flow: importing a template triggers a fresh review pass by the importing user, with the flagged-anomalies output prominently displayed, and the import cannot complete without explicit per-template approval.

This is a UX/flow decision, not a template-language migration. Jinja itself stays; what changes is the approval workflow at the import boundary.

The run.* namespace exposed in every template

run.id, run.started_at, run.automation_id, run.automation_name,
run.automation_version, run.trigger_type, run.trigger_id,
run.search_space_id, run.creator_id, run.attempt,
run.failed_step_id, run.error.*   (only in on_failure context)

Default value rendering

Non-string template values render as JSON by default (via the finalize hook): lists become ["a", "b"], dicts become {"k": "v"}, datetimes become ISO 8601. The | join, | length, | tojson filters give explicit control. Strings render as themselves with no quoting. None renders as empty string in templates, as null in JSON contexts.

7.5 Event bus

domain_events table, polled by Celery Beat alongside the existing scheduler. Both connector events and internal SurfSense events publish to it. Both are consumed by the dispatcher's event-trigger subscriber.

Automations themselves publish events. Successful and failed runs emit automation.run.succeeded / automation.run.failed events with the run metadata. This makes automations composable through events — chain them by subscribing one automation's event trigger to another's run event. No new mechanism; the trigger filter and event publishing already exist.

Upgrade path documented: when throughput or latency demands it, replace PostgreSQL polling with Redis Streams. The events.publish() and events.subscribe() interfaces stay the same. Nothing else changes.


8. Cross-cutting concerns

Concurrency policy

Per-automation concurrency field controls what happens when a new fire occurs while a previous run is still running:

  • drop_if_running — silently skip the new fire
  • queue — execute serially, in arrival order
  • allow_parallel — start a new run independently

The dispatcher enforces this before enqueueing.

Retry policy

Three fields, per-automation defaults with optional per-step overrides:

  • max_retries: integer, 010
  • retry_backoff: none | linear | exponential
  • timeout_seconds: integer

Retries on:

  • Action handler exceptions
  • Output schema validation failures (for dynamic-output actions, the validation error is fed back to the LLM in the retry)

Not retries:

  • when: evaluation failures (these are user errors, surface immediately)
  • Input validation failures (caught at dispatch, never reach the executor)

Budget enforcement (deferred — not in v1)

Future shape: budget_cap_usd on Execution, dispatcher refuses to enqueue if estimated cost exceeds it, executor kills the run if accumulated cost crosses it mid-flight (the LLM ops handler reports tokens consumed back to the executor between calls).

Prerequisites before this can land:

  • Each action declares cost reporting (tokens × model price, API call charges) — ActionDefinition has no such field today.
  • automation_runs.cost_usd column + executor accumulates per step.
  • A historical-cost ledger so pre-flight estimation can return useful numbers (otherwise the dispatcher gate is guessing).

Until all three exist, v1 has no surface for budget enforcement.

On-failure handlers

execution.on_failure is a list of steps that run after the main plan has failed and all retries are exhausted. Same step shape as the main plan. Cannot have their own on_failure. See run.error.* in the run context.

Artifacts

Actions that produce artifacts declare produces_artifacts: list[ArtifactSpec]:

@dataclass
class ArtifactSpec:
    kind: str           # "audio", "document", "image", "data"
    retention: str      # "transient" | "default" | "permanent"
    visibility: str     # "private" | "search_space" | "shared"

The engine handles storage (writes to SurfSense's existing object storage), URL generation (signed, scoped to the run's permissions), and cleanup (a nightly Celery Beat task deletes expired artifacts).

Duration classes and queue routing — deferred

The original design routed runs to multiple Celery queues based on each action's declared expected_duration_seconds. v1 ships with one queue (automations_default) and actions do not declare a duration. Multi-queue routing returns when burst load on a single queue actually justifies the operational complexity of independent worker pools.

Adding the second queue is a config change plus reintroducing expected_duration_seconds on the ActionDefinition dataclass — both mechanical, additive, and free of design rewrite.


9. Data model

v1 ships three tables: automations, automation_triggers, automation_runs. All scoped by search_space_id for RBAC.

The other three tables described in earlier drafts are deferred:

  • domain_eventsdeferred to Phase 3 (introduced with the event trigger).
  • mcp_connections, mcp_toolsdeferred to Phase 4 (MCP integration).

The deferred tables ship as-is when their consuming feature lands; nothing in the v1 schema needs to change to accommodate them. The three v1 tables form the engine's persistent state — definitions, triggers, and an immutable run history.

automations

field type notes
id int PK
search_space_id FK → search_spaces.id
created_by FK → users.id runs execute as this identity
name str
description str
status enum active, paused, archived
definition jsonb the editable structured spec
version int bumped on every edit
created_at / updated_at timestamps

automation_triggers

field type notes
id int PK
automation_id FK
type enum: schedule, manual (Phase 2/3 add webhook, event)
params jsonb trigger-type config, validated against trigger's params_schema
static_inputs jsonb per-attachment domain values merged into every run (static wins on collision)
enabled bool
last_fired_at timestamp
next_fire_at timestamp / null precomputed next fire moment for schedule triggers

secret_hash (for webhook bearer tokens) is deferred to Phase 2 with the webhook trigger.

automation_runs

field type notes
id int PK
automation_id FK
trigger_id FK / null null = manual via UI
status enum pending, running, succeeded, failed, cancelled, timed_out
definition_snapshot jsonb the definition as it was when this run fired
inputs jsonb merged & validated inputs (trigger.static_inputs producer runtime data, static wins)
step_results jsonb array of per-step results with timing
output jsonb / null
artifacts jsonb references to created artifacts
error jsonb / null
started_at / finished_at timestamps
agent_session_id str / null link to LangGraph trace if agent_task was used

cost_usd (per-run accumulated cost) is deferred until at least one action records token-level cost. When reintroduced it lands as a column-only migration.

Deferred tables

  • domain_events — the event bus backing event triggers. Ships in Phase 3 with the event trigger. v1 only emits automation.run.* events into application logs; the table is added when at least one consumer needs to subscribe to them.
  • mcp_connections / mcp_tools — see §3. Both ship in Phase 4 alongside the MCP harvester and the two-tier registry.

NL drafts are not a core table. They live in a generic short-TTL store (Redis or a transient table) when the NL flow is built in Phase 3.


10. NL authoring flow

This is how the system is intended to be used from day one, not just a Phase 3 addition. The product surface is: user describes intent in natural language, LLM produces a structured proposal, user reviews and edits in an auto-generated form, then saves. Hand-authoring JSON directly is supported but is not the primary path.

This shapes the trust model. Templates are LLM-generated from day one, not hand-written by power users. The mitigation is human-in-the-loop review, not "trusted authors only."

Pass 1: Proposal generation

User provides natural-language input. The Generator LLM is given:

  • The full schema set (input schema for definition, registry of action types with their params_schemas, registry of trigger types, list of allowed Jinja filters)
  • A tool to list available connectors, channels, and other SearchSpace resources, so it doesn't invent names that don't exist
  • A few-shot set of examples

Scoped input. The Generator does not receive arbitrary SearchSpace document content. Its context is the user's prompt plus the schema and registry information. This minimizes the prompt-injection surface — there's no document text in the context for an attacker to seed instructions into.

If a user wants document-aware generation later ("create an automation that processes documents like this one"), that's a deliberate feature extension with its own prompt-injection mitigations, not the default flow.

Output: a structured proposal matching the automation definition schema.

Pass 2: Deterministic validation

Server-side, before the proposal reaches the user:

  • Validate against JSON Schema (shape correctness)
  • Verify every action and trigger type referenced exists in the registry
  • Verify every connector/channel/resource referenced exists in this SearchSpace
  • Validate every template against the sandbox's allowlist (no underscore attributes, no unregistered filter names, length under cap)

Failures here are deterministic errors, not warnings. A proposal that references a non-existent action or includes a template using {{x.__class__}} is rejected before the user sees it; the Generator is re-prompted with the validation error and asked to fix the proposal.

Pass 2.5: Review pass

A second LLM call — the Review LLM — examines the validated proposal and produces two outputs for the user:

  1. A plain-language summary of what the automation will do, in business terms. "This automation will run every weekday at 9am. It reads documents in this SearchSpace tagged 'competitor' that were indexed since the last run, asks an agent to summarize them as 5 bullets, and posts the summary to your #engineering-standup Slack channel. Estimated cost: $0.40 per run."

  2. A "things worth checking" list flagging anything unusual:

    • Templates with unusual attribute paths or filter usage
    • Prompts containing instructions that look more like commands than descriptions ("ignore previous instructions" style)
    • Action sequences that touch external systems without obvious benefit to the user
    • Cost estimates that seem high relative to the goal
    • References to actions the user hasn't used before
    • Schedules tighter than 15 minutes (likely should be event triggers)

The Review LLM is a UX layer that makes review actually useful. It is not a security boundary. The deterministic controls (sandbox, runtime limits, schema validator) are the security boundaries. The Review LLM helps users catch their own intent mismatches and surfaces anomalies for attention, but the sandbox would block dangerous templates even if the Review LLM missed them.

This separation is important: two probabilistic controls compounding can create a false sense of security. The Review LLM is explicitly framed in the architecture as helper, not gatekeeper.

Pass 3: Editable review

The user lands on a form pre-filled with the proposal. The page shows:

  • The plain-language summary from the Review pass
  • The flagged items, prominently displayed near the relevant fields
  • The full editable form, auto-generated from the JSON Schemas
  • Cost estimate and impact summary (which external systems get touched)

Every field is editable. Clarifications appear as required fields. Templates are shown in code-styled fields with syntax highlighting and the filter palette visible. The user can edit any field; saving re-runs Pass 2 (deterministic validation) before persisting.

Hitting Save promotes the proposal to an automation row.

Editing existing automations

NL editing of an existing automation is a patch operation: the Generator LLM receives the current definition plus the NL instruction and produces a modified proposal. The same Pass 2 (validation) and Pass 2.5 (review) run against the modified version, and the user reviews the diff before saving. Existing run history is unaffected — only future runs use the new version.

Why human-in-the-loop is non-negotiable

The Generator LLM, the Review LLM, and the sandbox are three layers of defense against malformed or malicious proposals. The human approval step is the fourth and most important layer. It exists because:

  • LLMs can be prompt-injected; humans can spot text that asks them to ignore instructions
  • LLMs can produce confident-but-wrong proposals; humans can catch semantic mismatches between intent and output
  • The cost of a bad automation running unattended is high; the cost of a user clicking "approve" after reading is low

The architecture must never offer "auto-approve" or "skip review" options for LLM-generated proposals. Save requires human action on the proposal, always.


11. Repository layout

surfsense_backend/app/
├── automations/                       # NEW: the engine
│   ├── __init__.py
│   ├── persistence/                   # SQLAlchemy models + enums for 3 tables
│   ├── schemas/                       # Pydantic schemas (definition envelope, etc.)
│   ├── routes.py                      # FastAPI router (/api/v1/automations)
│   ├── service.py                     # CRUD + business logic
│   ├── dispatcher.py                  # trigger matching, run creation
│   ├── executor.py                    # the Celery task that runs a plan
│   ├── templating.py                  # Jinja sandbox + filters
│   ├── events.py                      # publish/subscribe for domain_events
│   ├── filters.py                     # JSON filter grammar evaluator
│   ├── registries/                    # action and trigger registries
│   │   ├── actions/                   # ActionDefinition + handler registration
│   │   └── triggers/                  # TriggerDefinition
│   └── nl/                            # Phase 1 — primary user path
│       ├── generator.py               # Generator LLM
│       ├── reviewer.py                # Review LLM (summary + flagged items)
│       ├── validator.py               # deterministic schema + resource checks
│       └── prompts.py                 # system prompts for both LLMs
│
├── utils/
│   └── periodic_scheduler.py          # EXTENDED to scan automation_triggers
│
└── alembic/versions/
    └── NN_add_automation_tables.py

surfsense_web/app/(routes)/
└── automations/                       # NEW: UI
    ├── page.tsx                       # list
    ├── new/page.tsx                   # NL input + draft preview (Phase 1)
    ├── [id]/page.tsx                  # editor (auto-generated forms)
    └── [id]/runs/page.tsx             # run history, streamed via Electric SQL

12. Phased delivery

Each phase delivers something usable. Each de-risks the next. NL authoring is the primary user path from Phase 1 — what evolves across phases is which actions and triggers are available, not whether users can describe automations in natural language.

Phase 1 — Engine MVP with NL authoring

Step 1 (current scope, this batch of commits):

  • 3 tables (automations, automation_triggers, automation_runs) + Alembic migration
  • Empty action and trigger registries under app/automations/registries/ (concrete entries land in later steps)
  • Pydantic schemas for the automation definition envelope, the two v1 trigger params shapes (schedule, manual), and the one v1 action params shape (agent_task)
  • Module structure under app/automations/ (persistence/, schemas/, registries/), fully isolated from the existing codebase

Step 2:

  • The agent_task action handler and the schedule / manual triggers registered in app/automations/registries/. Tool resolution for agent_task.params.tools is opaque to the contract — the handler decides what string identifiers it accepts and how they resolve.

Step 3:

  • Executor (single-queue Celery task) with retries and timeouts
  • Template engine (Jinja sandbox + the v1 filter allowlist + runtime limits)
  • Manual "Run now" endpoint

Step 4:

  • NL authoring flow: Generator LLM, deterministic validator, Review LLM, editable form
  • Run history UI with Electric SQL streaming

After Phase 1: a user can describe an automation in natural language, review the proposal (with summary + flagged anomalies), edit any field, save, and watch it run on a schedule.

Phase 2 — Webhooks and delivery

  • webhook trigger with per-automation bearer tokens
  • Tight actions: slack_post, send_email, notification
  • transform_data action
  • on_failure hooks
  • Step-level retry/timeout overrides
  • Concurrency policy enforcement

After Phase 2: external systems can drive automations, results go somewhere humans see, complex pipelines have proper error handling.

Phase 3 — NL authoring polish

  • NL patch flow for editing existing automations (diff-based)
  • Conversational refinement during proposal review ("change the schedule to weekdays only," "add a Slack notification on failure")
  • Improved Review LLM coverage (more anomaly patterns, cost-relative-to- goal heuristics)
  • Saved prompt templates and starter examples

After Phase 3: NL authoring is the polished primary surface; edit flows are conversational rather than form-only.

Phase 4 — Event triggers + integration tooling

  • domain_events table and events.py module
  • Indexing pipeline publishes connector.* events (smallest change — just add publish calls to the existing flow)
  • Automations publish automation.run.* events on completion
  • event trigger with filter grammar
  • The unification layer redesign (see §3) — CallContext, scope declarations, per-user authorization gating
  • MCP integration on top of the unification layer (external tool servers harvested into the shared catalog)

After Phase 4: "do X when Y happens" automations work, including automation-chaining through events; external MCP tools and SurfSense actions share one vocabulary.

Phase 5 — Wrapping existing features and sharing

  • Wrap existing SurfSense features as actions: podcast_generation, report_generation, indexing_sweep
  • Artifact lifecycle implementation
  • expected_duration_seconds based queue routing (split automations_long from automations_default)
  • Automation templates (shareable, exportable, importable) — with the import re-approval flow that handles the approver-≠-runner trust shift documented in §7.4's pre-Phase-5 gate
  • Cross-automation composition examples in the docs

After Phase 5: every existing SurfSense feature is automatable without any per-feature code, and automations can be shared between SearchSpaces and users.


13. Decisions locked

For reference — every decision made through the design process, in one place.

Foundations

  1. JSON Schema (draft 2020-12) is the single schema language for everything
  2. Definition is the program; infrastructure is the interpreter
  3. List of steps (not single action) in the plan, with output_as chaining
  4. ⏸ Capability unification layer (one catalog shared by automations, agents, and future surfaces) — deferred to post-v1 (see §3). v1 ships actions only.
  5. Name-based resolution: definitions reference action and trigger types by string ID. The registry is the runtime's vocabulary; lookup is a dict access. No code references in definitions.
  6. The expressive spectrum runs from pure direct calls to broad agent_task; the NL generator proposes the cheapest shape that meets intent (Shape 6 from §4 by default)

Trigger taxonomy

  1. Three trigger types: schedule, webhook, event
  2. Events absorb both connector events and internal SurfSense events
  3. Filter grammar is JSON-structured operators (not Jinja)

Templating cluster

  1. Jinja2 SandboxedEnvironment for templates and when: predicates — but with the explicit understanding that the sandbox is an allowlist-by-default architecture, not a denylist
  2. Zero globals registered. Curated 15 filters only, each audited for safe behavior with hostile input. List grows only by reviewed addition
  3. Four runtime mitigations: StrictUndefined, 8 KB template source cap, 100 ms render time cap (watchdog-enforced), 1 MB output size cap
  4. Non-string template values render as JSON by default
  5. Fixed run.* namespace, documented
  6. Pre-Phase-5 gate: template sharing across SearchSpaces breaks the approver-equals-runner trust model. Mitigation is a re-approval flow at the import boundary (UX-level), not a template-language migration. Jinja itself stays.

Execution

  1. Executor is a Celery task wrapping a sequential loop — not an agent
  2. when: is optional per step; false = skipped (not failed)
  3. No DAGs, no parallelism, no loops — composition via agent_task or events
  4. on_failure part of execution policy from v1
  5. Step-level retry and timeout overrides
  6. ⏸ Budget cap enforced pre-enqueue and mid-flight — deferred until the cost ledger ships (see §8 Budget enforcement)

Components

  1. Dispatcher / executor / handlers / registry — distinct, each replaceable
  2. ⏸ Side effects are a set, including USER_VISIBLEdeferred until multi-user automation RBAC ships
  3. expected_duration_seconds integer drives queue routing — deferred until a second Celery queue is needed
  4. produces_artifacts is a list of ArtifactSpec, not a bool — deferred until artifacts beyond the deliverable handlers' own persistence are needed
  5. Output schemas recommended on agent_task; editor warns when missing

Event bus

  1. domain_events table for v1, with upgrade path to Redis Streams
  2. Automations publish run events for composability
  3. Publish/subscribe behind interface — no direct table access elsewhere

Capability unification — all deferred to post-v1

  1. ⏸ One shared catalog of "things this SurfSense instance can do" — deferred, see §3
  2. ⏸ Handler CallContext (caller user id, search space id, run id) — deferred with unification
  3. ⏸ Per-capability scope declarations driving authorization — deferred with unification
  4. ⏸ MCP integration on top of the unification layer (mcp_connections, mcp_tools, harvester) — deferred to Phase 4

Credentials — all deferred to Phase 2

  1. ⏸ Credentials never appear in the automation definition — only connection IDs do — Phase 2
  2. ⏸ Credentials never appear in the LLM's context — the host holds them — Phase 2
  3. ⏸ Credentials resolved per-call by the handler context, not pre-loaded into worker environment — Phase 2
  4. ⏸ Tokens encrypted at rest; refresh handled automatically by the handler context — Phase 2

v1-minimum

  1. v1 ships actions only — no separate capability layer. ActionDefinition is five fields: type, name, description, params_schema, handler. Additional fields are added only when a concrete consumer feature requires them.
  2. Cost is measured from a per-run ledger, not declared. Pre-flight cost checks return when the ledger has enough history.
  3. Single automations_default Celery queue in v1. Multi-queue routing returns when load justifies it.

NL authoring

  1. LLM-authored templates is the primary path from day one — not a Phase 3 addition. Hand-authoring JSON is supported but secondary
  2. Generator LLM produces JSON; deterministic schema + resource validation runs before user sees the proposal
  3. Review LLM produces plain-language summary + flagged anomalies for the user — UX layer, not a security boundary
  4. Generator LLM's input is scoped (user prompt + schema + registry only); arbitrary document content is not fed in
  5. Human approval is required before save — no auto-approval option, ever
  6. Every field editable in the proposal; unresolved questions surface as clarifications
  7. NL drafts are transient storage, not a core table

Data model

  1. v1 ships three tables (automations, automation_triggers, automation_runs). domain_events lands in Phase 3; mcp_connections and mcp_tools in Phase 4.
  2. Run rows snapshot the definition (immutable history)
  3. All entities scoped by search_space_id for RBAC
  4. Editing an automation bumps version; existing runs unaffected

14. Open questions deferred to implementation

None of these block design; they're decisions a developer will make in context, with the principle from §1 as their guide.

  • Exact retry backoff formulas (multipliers, jitter, ceilings)
  • Webhook signature verification standards (HMAC scheme, header naming)
  • Whether to support inline JSON Schema $ref to external schemas, or inline everything
  • Specific CDN/storage backend choices for artifacts (probably whatever SurfSense already uses for podcasts)
  • Rate limits per SearchSpace and per user
  • Audit log retention policy

15. Why this is ready to build

This document satisfies five tests:

  1. The four worked examples (digest, CI webhook, file-added-trigger, weekly podcast) all express cleanly in the contract without special cases. Each one was used to find gaps before the gaps reached code.

  2. The audit pass identified six refinements, all incorporated. No pending audit items.

  3. Every decision points back to the principle from §1. When a future feature request lands, "does it belong in the definition or in the engine?" gives a clear answer.

  4. The build is staged so Phase 1 ships in weeks, not months, and each subsequent phase delivers user value while de-risking the next.

  5. Existing SurfSense infrastructure is reused, not paralleled. Celery Beat, PostgreSQL/JSONB, Electric SQL, SQLAlchemy/Alembic, the existing tools/registry.py pattern, the existing Search Space scoping — all continue to do what they already do. The automation engine is a new directory, not a new system.

The next document a developer needs is the Pydantic models and JSON Schemas spelled out concretely. Those follow mechanically from this plan.


Sources consulted: Claude Code Routines documentation; NousResearch/hermes- agent (cron and skills subsystems); n8n documentation on node types and workflow data model; the SurfSense repository and DeepWiki architecture notes (FastAPI + Celery Beat + Electric SQL + LangGraph Deep Agents + Search Space RBAC); Model Context Protocol specification for external tool harvesting; AWS EventBridge for filter grammar; workflow-pattern literature (van der Aalst et al.) for the trigger / action / concurrency vocabulary.