Agentic AI Orchestration Platform
Multi-agent runtime on a plan-retrieve-act-critique loop, emitting a full decision and tool-call trace on every run.
Use case
The problem space.
Make knowledge-work requests repeatable and auditable — the reasoning trace and tool activity stay visible rather than collapsing into an answer.
What's implemented
Built with intention.
Specialised agent loop01
Planner, retriever, executor, and critic agents share task state and take turns through explicit tool-calling contracts.
Safe recovery02
A sandboxed execution loop captures failed actions, lets the critic propose corrections, and bounds retries with policy checks.
Governance surface03
WebSocket events stream live traces to React; PostgreSQL persists inputs, tools, outputs, and checkpoints for replay.
Architecture
Systems in concert.
The primary request and data paths, presented as a compact operating model.
A conceptual architecture for communicating the system design and operational responsibilities.
Step by step
From zero to
running.
Representative local-development commands that show the implementation path and operating sequence.
- 01
Configure model and storage
Copy the example environment file and provide model, vector-store, and database credentials.
cp .env.example .env docker compose up -d postgres weaviate - 02
Run the orchestration API
Install Python dependencies, apply the schema, and start the FastAPI development server.
uv sync uv run alembic upgrade head uv run fastapi dev app/main.py - 03
Open the trace console
Launch the React client and connect it to the WebSocket endpoint exposed by the API.
cd web && npm install && npm run dev - 04
Evaluate a task suite
Run the Ragas-style harness against the pinned benchmark cases and compare success rate by agent stage.
uv run python -m evals.run --suite core-150