trustgraph/trustgraph-cli/trustgraph/cli/load_doc_embeds.py
cybermaggedon 89be656990
Release/v1.2 (#457)
* Bump setup.py versions for 1.1

* PoC MCP server (#419)

* Very initial MCP server PoC for TrustGraph

* Put service on port 8000

* Add MCP container and packages to buildout

* Update docs for API/CLI changes in 1.0 (#421)

* Update some API basics for the 0.23/1.0 API change

* Add MCP container push (#425)

* Add command args to the MCP server (#426)

* Host and port parameters

* Added websocket arg

* More docs

* MCP client support (#427)

- MCP client service
- Tool request/response schema
- API gateway support for mcp-tool
- Message translation for tool request & response
- Make mcp-tool using configuration service for information
  about where the MCP services are.

* Feature/react call mcp (#428)

Key Features

  - MCP Tool Integration: Added core MCP tool support with ToolClientSpec and ToolClient classes
  - API Enhancement: New mcp_tool method for flow-specific tool invocation
  - CLI Tooling: New tg-invoke-mcp-tool command for testing MCP integration
  - React Agent Enhancement: Fixed and improved multi-tool invocation capabilities
  - Tool Management: Enhanced CLI for tool configuration and management

Changes

  - Added MCP tool invocation to API with flow-specific integration
  - Implemented ToolClientSpec and ToolClient for tool call handling
  - Updated agent-manager-react to invoke MCP tools with configurable types
  - Enhanced CLI with new commands and improved help text
  - Added comprehensive documentation for new CLI commands
  - Improved tool configuration management

Testing

  - Added tg-invoke-mcp-tool CLI command for isolated MCP integration testing
  - Enhanced agent capability to invoke multiple tools simultaneously

* Test suite executed from CI pipeline (#433)

* Test strategy & test cases

* Unit tests

* Integration tests

* Extending test coverage (#434)

* Contract tests

* Testing embeedings

* Agent unit tests

* Knowledge pipeline tests

* Turn on contract tests

* Increase storage test coverage (#435)

* Fixing storage and adding tests

* PR pipeline only runs quick tests

* Empty configuration is returned as empty list, previously was not in response (#436)

* Update config util to take files as well as command-line text (#437)

* Updated CLI invocation and config model for tools and mcp (#438)

* Updated CLI invocation and config model for tools and mcp

* CLI anomalies

* Tweaked the MCP tool implementation for new model

* Update agent implementation to match the new model

* Fix agent tools, now all tested

* Fixed integration tests

* Fix MCP delete tool params

* Update Python deps to 1.2

* Update to enable knowledge extraction using the agent framework (#439)

* Implement KG extraction agent (kg-extract-agent)

* Using ReAct framework (agent-manager-react)
 
* ReAct manager had an issue when emitting JSON, which conflicts which ReAct manager's own JSON messages, so refactored ReAct manager to use traditional ReAct messages, non-JSON structure.
 
* Minor refactor to take the prompt template client out of prompt-template so it can be more readily used by other modules. kg-extract-agent uses this framework.

* Migrate from setup.py to pyproject.toml (#440)

* Converted setup.py to pyproject.toml

* Modern package infrastructure as recommended by py docs

* Install missing build deps (#441)

* Install missing build deps (#442)

* Implement logging strategy (#444)

* Logging strategy and convert all prints() to logging invocations

* Fix/startup failure (#445)

* Fix loggin startup problems

* Fix logging startup problems (#446)

* Fix logging startup problems (#447)

* Fixed Mistral OCR to use current API (#448)

* Fixed Mistral OCR to use current API

* Added PDF decoder tests

* Fix Mistral OCR ident to be standard pdf-decoder (#450)

* Fix Mistral OCR ident to be standard pdf-decoder

* Correct test

* Schema structure refactor (#451)

* Write schema refactor spec

* Implemented schema refactor spec

* Structure data mvp (#452)

* Structured data tech spec

* Architecture principles

* New schemas

* Updated schemas and specs

* Object extractor

* Add .coveragerc

* New tests

* Cassandra object storage

* Trying to object extraction working, issues exist

* Validate librarian collection (#453)

* Fix token chunker, broken API invocation (#454)

* Fix token chunker, broken API invocation (#455)

* Knowledge load utility CLI (#456)

* Knowledge loader

* More tests
2025-08-18 20:56:09 +01:00

233 lines
5.5 KiB
Python

"""
This utility takes a document embeddings core and loads it into a running
TrustGraph through the API. The document embeddings core should be in msgpack
format, which is the default format produce by tg-save-doc-embeds.
"""
import aiohttp
import asyncio
import msgpack
import json
import sys
import argparse
import os
import signal
class Running:
def __init__(self): self.running = True
def get(self): return self.running
def stop(self): self.running = False
de_counts = 0
async def load_de(running, queue, url):
global de_counts
async with aiohttp.ClientSession() as session:
async with session.ws_connect(url) as ws:
while running.get():
try:
msg = await asyncio.wait_for(queue.get(), 1)
# End of load
if msg is None:
break
except:
# Hopefully it's TimeoutError. Annoying to match since
# it changed in 3.11.
continue
msg = {
"metadata": {
"id": msg["m"]["i"],
"metadata": msg["m"]["m"],
"user": msg["m"]["u"],
"collection": msg["m"]["c"],
},
"chunks": [
{
"chunk": chunk["c"],
"vectors": chunk["v"],
}
for chunk in msg["c"]
],
}
try:
await ws.send_json(msg)
except Exception as e:
print(e)
de_counts += 1
async def stats(running):
global de_counts
while running.get():
await asyncio.sleep(2)
print(
f"Graph embeddings: {de_counts:10d}"
)
async def loader(running, de_queue, path, format, user, collection):
if format == "json":
raise RuntimeError("Not implemented")
else:
with open(path, "rb") as f:
unpacker = msgpack.Unpacker(f, raw=False)
while running.get():
try:
unpacked = unpacker.unpack()
except:
break
if user:
unpacked["metadata"]["user"] = user
if collection:
unpacked["metadata"]["collection"] = collection
if unpacked[0] == "de":
qtype = de_queue
while running.get():
try:
await asyncio.wait_for(qtype.put(unpacked[1]), 0.5)
# Successful put message, move on
break
except:
# Hopefully it's TimeoutError. Annoying to match since
# it changed in 3.11.
continue
if not running.get(): break
# Put 'None' on end of queue to finish
while running.get():
try:
await asyncio.wait_for(de_queue.put(None), 1)
# Successful put message, move on
break
except:
# Hopefully it's TimeoutError. Annoying to match since
# it changed in 3.11.
continue
async def run(running, **args):
# Maxsize on queues reduces back-pressure so tg-load-kg-core doesn't
# grow to eat all memory
de_q = asyncio.Queue(maxsize=10)
url = args["url"]
flow_id = args["flow_id"]
load_task = asyncio.create_task(
loader(
running=running,
de_queue=de_q,
path=args["input_file"], format=args["format"],
user=args["user"], collection=args["collection"],
)
)
de_task = asyncio.create_task(
load_de(
running=running,
queue=de_q,
url = f"{url}api/v1/flow/{flow_id}/import/document-embeddings"
)
)
stats_task = asyncio.create_task(stats(running))
await de_task
running.stop()
await load_task
await stats_task
async def main(running):
parser = argparse.ArgumentParser(
prog='tg-load-doc-embeds',
description=__doc__,
)
default_url = os.getenv("TRUSTGRAPH_API", "http://localhost:8088/")
default_user = "trustgraph"
collection = "default"
parser.add_argument(
'-u', '--url',
default=default_url,
help=f'TrustGraph API URL (default: {default_url})',
)
parser.add_argument(
'-f', '--flow-id',
default="default",
help=f'Flow ID (default: default)'
)
parser.add_argument(
'-i', '--input-file',
# Make it mandatory, difficult to over-write an existing file
required=True,
help=f'Output file'
)
parser.add_argument(
'--format',
default="msgpack",
choices=["msgpack", "json"],
help=f'Output format (default: msgpack)',
)
parser.add_argument(
'--user',
help=f'User ID to load as (default: from input)'
)
parser.add_argument(
'--collection',
help=f'Collection ID to load as (default: from input)'
)
args = parser.parse_args()
await run(running, **vars(args))
running = Running()
def interrupt(sig, frame):
running.stop()
print('Interrupt')
signal.signal(signal.SIGINT, interrupt)
asyncio.run(main(running))