SourceLogic SaaS (Multi-Tenant AI RAG)

Local-first RAG assistant for repository exploration. The system indexes source files into ChromaDB and serves grounded responses through a React UI and FastAPI backend.

Stack

• Frontend: React + Vite + TypeScript
• Backend: FastAPI + SQLAlchemy (SQLite) + LangChain
• Retrieval Store: ChromaDB
• LLM: OpenAI via LangChain adapters

Architecture Diagram

flowchart LR
    U[User] --> UI[React UI :5173]
    UI --> API[FastAPI :8000]
    API --> Ingest[Ingest Service]
    API --> Chat[Chat Service]
    Ingest --> CH[(ChromaDB)]
    Chat --> CH
    Chat --> OAI[OpenAI]

Loading

Setup

Backend (secrets only)

cd Chat-Project\backend
python -m venv .venv
.\.venv\Scripts\Activate.ps1
pip install -r requirements.txt
Copy-Item .env.example .env

backend/.env must contain:

OPENAI_API_KEY=your_openai_api_key
DATABASE_URL=sqlite+aiosqlite:///./data/codechat.db
CHROMA_PATH=./data/chroma_db

Run backend on port 8000 (from Chat-Project root):

cd Chat-Project
python -m uvicorn backend.app.main:app --reload --host 0.0.0.0 --port 8000

Frontend (public config only)

cd Chat-Project\frontend
npm install
Copy-Item .env.example .env

frontend/.env must contain only:

VITE_API_URL=http://localhost:8000

Run frontend on port 5173:

cd Chat-Project\frontend
npm run dev -- --host 0.0.0.0 --port 5173

No root-level .env is used.

API Contract (Real vs Planned)

Endpoint	Method	Status	Notes
/workspaces	GET	Real	List registered workspaces
/workspaces	POST	Real	Register workspace root path
/workspaces/{workspace_id}/status	GET	Real	Workspace indexing status
/workspaces/{workspace_id}/sessions	POST	Real	Create chat session
/sessions/{session_id}/history	GET	Real	Retrieve chat history
/workspaces/{workspace_id}/ingest (/ingest)	POST	Real	Triggers background codebase ingestion to ChromaDB
/chat/{session_id}/stream (/stream)	POST	Real	Stateful streaming LLM response using DB-backed Chat Memory

Known Limitations

• Real authentication flow (OAuth2/JWT) is currently mocked via X-Tenant-ID header to demonstrate multi-tenant data isolation purely at the storage and logic layer.

Features

• Hybrid Architecture (Local RAG + OpenAI): Processes code embeddings locally with open-source models while selectively querying GPT-4o for complex reasoning.
• SaaS-Grade Multi-Tenant Isolation: Built with scalable Data Isolation principles, enforcing strictly separate contexts (X-Tenant-ID) across Workspaces, Sessions, and Vector Store chunks.
• Agentic Polling & Streaming: Real-time feedback via SSE (Server-Sent Events) for the chat and asynchronous background tasks for heavy ingests.

Trade-offs

• frontend/.env is intentionally limited to public runtime config (VITE_API_URL) to avoid secret leakage in browser bundles.
• Secrets are kept in backend/.env for local development simplicity; enterprise deployments should move them to a managed secret store (Vault/KMS).

Security Notice

• Critical To-Do: if any real OPENAI_API_KEY was ever committed or shared from the previous root .env, rotate it immediately.

Project Structure

backend/
  .env               # secrets (not committed)
  .env.example       # placeholders only
  app/
    api/v1/          # FastAPI routes
    core/            # config + database setup
    models/          # SQLAlchemy models
    schemas/         # Pydantic models
    services/        # RAG + ingestion + chat logic
frontend/
  .env               # public UI config only
  .env.example
  src/               # React application

## Credits

Developed and maintained by **Federico Orsi**.