Relay · Agent Platform

An agent is more than a prompt: it needs credentials, tool permissions, a schedule, execution history, and a way to report its result. Relay brings those pieces into a single application for developers exploring how to operate agent workflows.

• Agent configuration, a conversational builder, and run history.
• Connections for services such as Gmail, Slack, Sheets, YouTube, and HTTP tools.
• MCP server configuration and tool discovery.
• Scheduled runs, worker execution, and notification dispatch.
• Team conversations with a lead agent and shared task context.
• Optional Jenkins lifecycle hooks and a text-only, synchronous A2A interface.

Use Node.js/npm, Python 3.12, and Docker Compose. Open a downloaded source archive in your terminal.

Backend — from the repository root:

cd backend
python3.12 -m venv .venv
./.venv/bin/pip install -r requirements.txt
docker compose up -d postgres redis
./.venv/bin/alembic upgrade head
bash scripts/dev.sh

The script starts the API, scheduler, worker, and notifier. API documentation is at localhost:8000/docs.

Frontend — in another terminal at the repository root:

npm install
npm run dev

Open localhost:3100. The package scripts explicitly select port 3100.

Backend settings are defined in backend/app/config.py and read from environment variables or backend/.env.

The default model provider is a fake provider for local development. Real model and connector calls require their own configuration. Demo authentication and development credential defaults are intended for local use.

flowchart LR
    UI[Next.js workspace] --> API[FastAPI routes]
    API --> DB[(SQL database)]
    API --> Q[(Redis queues)]
    S[Scheduler] --> Q
    Q --> W[Agent worker]
    W --> M[Model provider]
    W --> T[MCP and connectors]
    W --> DB
    W --> N[Notification queue]
    N --> D[Notification dispatcher]

The API manages configuration and requests. Background processes handle scheduled execution and delivery, while SQL records agents and runs. Redis is needed for queues shared between the separate processes started by dev.sh; the in-process fallback is intended for single-process use.

Jenkins deployment is an integration hook; its deployment step is a placeholder. A2A supports discovery and synchronous text SendMessage, not every protocol capability. Consult the backend guide and flow documentation before enabling external integrations.

Explore more of my work: latest open-source tools · recent projects.