rachel-proxy 0.1.0a1

RACHEL (Rpg Agent CHat Evaluation Loop) - OpenAI-compatible Chat Completion API proxy with stateful LangGraph agent and V8 code sandbox

---

title: RACHEL (rachel-proxy) emoji: 🎲 colorFrom: red colorTo: blue sdk: docker app_port: 7860 pinned: false

RACHEL (rachel-proxy)

RACHEL (Rpg Agent CHat Evaluation Loop) is a FastAPI proxy that sits between JanitorAI (or any OpenAI-compatible client) and LLM completion providers (OpenRouter, OpenAI, Google Gemini, DeepSeek), running request payloads through a stateful LangGraph agent with a secure Python/V8 code sandbox and dice rolling tools.

• Why Use RACHEL? — Core features, benefits, assumptions, and design philosophies.
• All About Sessions — How session IDs are resolved and managed via API endpoints.
• Road to Multi-Tenant — Multi-tenant cloud roadmap and architecture.

---

One-Click Desktop Launchers (Single-Tenant Mode)

Download the release zip for your operating system from Releases and launch with one click:

Windows

1. Unzip rachel-proxy-vX.X.X-win-x64.zip.
2. Double-click launch.bat.
3. Security Warning Bypass: If Windows SmartScreen displays "Windows protected your PC", click More info $\rightarrow$ Run anyway.

macOS

1. Unzip rachel-proxy-vX.X.X-mac-arm64.zip (Apple Silicon) or rachel-proxy-vX.X.X-mac-x64.zip (Intel).
2. Double-click launch.command.
3. Security Warning Bypass: If macOS blocks execution ("Unidentified Developer"), open System Settings $\rightarrow$ Privacy & Security $\rightarrow$ click Open Anyway, or run xattr -cr launch.command in Terminal.

Linux

1. Unzip rachel-proxy-vX.X.X-linux-x64.zip.
2. Double-click launch.sh (or rachel-proxy.desktop).

---

Multi-Tenant Cloud Deployment (GCP Cloud Run + Neon PostgreSQL)

Deploy RACHEL to GCP Cloud Run using the multi-stage Dockerfile:

1. Database Provisioning

Run scripts/schema_v1.sql against your Neon PostgreSQL database instance to initialize tables.

2. Build & Push Docker Container

docker build -t gcr.io/YOUR_PROJECT_ID/rachel-proxy:latest .
docker push gcr.io/YOUR_PROJECT_ID/rachel-proxy:latest

3. Deploy to GCP Cloud Run

Configure the container with environment variables (using GCP Secret Manager for sensitive keys):

gcloud run deploy rachel-proxy \
  --image gcr.io/YOUR_PROJECT_ID/rachel-proxy:latest \
  --platform managed \
  --region us-central1 \
  --set-env-vars MULTI_TENANT_MODE=true \
  --set-env-vars DATABASE_URL="postgresql://user:[e-mail]/neondb?sslmode=require" \
  --set-env-vars ENCRYPTION_MASTER_KEY="your-secure-master-encryption-secret" \
  --set-env-vars OIDC_ISSUER_URL="https://your-tenant.clerk.accounts.dev" \
  --set-env-vars OIDC_JWKS_URL="https://your-tenant.clerk.accounts.dev/.well-known/jwks.json" \
  --allow-unauthenticated

---

Developer Local Setup

If you are cloning this repository for local development:

1. Install Python dependencies:

   pip install -e .
   

2. Build Frontend Assets (Required):

   [!IMPORTANT] The compiled static frontend directory (src/rachel/static/) is git-ignored. You MUST build the frontend before launching the backend server:

   python scripts/build_frontend.py --target local
   

3. Run the development server:

   uvicorn rachel.proxy:app --reload --host 0.0.0.0 --port 8000
   

---

Initial Setup & LLM Provider Credentials

Once the proxy starts, open the Admin Console in your browser at http://localhost:8000:

1. Proxy API Key: Enter the local admin key (printed to console logs or saved in data/proxy.key).
2. Provider Credentials: Configure your preferred provider (OpenRouter BYOK / PKCE, OpenAI, Google Gemini, or DeepSeek) directly in the Provider Credentials card.
3. Select your Active Provider and save settings.

Captured payloads are appended to docs/example-janitorai-payload.md.