CONTEXT.md

CONTEXT

FinSight is a Python-first, spec-driven system.

Environment

• Development on Windows laptop
• Deployment on Ubuntu server
• Local model endpoint available on LAN
• Offline-capable tests and tooling
• WSL2 may expose Docker as either docker (native Linux CLI) or docker.exe (Windows bridge)
• Repo wrapper scripts/docker-auto.sh is the standard automation entrypoint for Docker commands

Architecture Constraints

• FastAPI backend
• LangGraph supervisor orchestration
• Celery workers with Redis
• PostgreSQL + pgvector
• Runtime behavior in config/runtime/*.yaml
• Config validation via Pydantic at startup (fail-fast)

Non-Negotiables

• No hardcoded business behavior in code
• No secrets in repository
• Strict typing with mypy --strict
• Lint with ruff
• Tests must run offline
• Agent role boundaries are enforced

Delivery Workflow

1. Read constitution and spec artifacts
2. Implement only files in the active plan
3. Run mypy, ruff, and pytest
4. Commit with conventional commit format