Agents Gateway
Open-source FastAPI gateway for AI agents — serve, orchestrate, and deploy production AI agents on top of Agno. MIT licensed.
The problem
Most production AI agents start as Python scripts that work fine in a notebook. Shipping them to production is a different problem entirely: API auth, per-user OAuth tokens, prompt versioning, multi-tenant isolation, job queues, approval flows, observability. Building all of that for every project doesn’t scale.
Agents Gateway is the runtime that handles those layers, so your agent code stays focused on the agent.
Why Agents Gateway?
A focused production scaffold around the Agno agent framework.
REST API for agents
Create, configure, version, and chat with agents over HTTP. Per-user sessions, streaming or non-streaming, multi-tenant by design.
Multi-agent teams
Compose agents into teams with coordinate or supervisor execution modes. Job queues, approval flows, containerized workers.
OAuth toolkits
Plug Gmail, Calendar, Contacts, and Drive (Google + Microsoft) into any agent with confirmation workflows and auto-refresh.
Production-ready ops
OpenTelemetry tracing, Sentry, OTLP, Logtail. Versioned prompts. Per-tenant knowledge base on Qdrant.
One-click deploy
Render, Railway, Koyeb buttons backed by a prebuilt multi-arch image on GHCR. K8s manifests + AWS/Azure/GCP recipes included.
Open by default
MIT licensed. No vendor lock-in. Bring your own model provider (OpenAI, Anthropic, Gemini, …) and your own database.
Quickstart — deploy in 5 minutes
git clone https://github.com/liberzon/agents-gateway
cd agents-gateway
# Start PostgreSQL + Qdrant (seeds demo agents automatically)
docker compose up -d
# Set up Python environment + start the API
./scripts/dev_setup.sh && source .venv/bin/activate
./scripts/start_server.sh
Then chat with the demo agent:
export GOOGLE_API_KEY="your-gemini-key"
curl -X POST http://localhost:8000/v2/agents/demo-assistant/chat \
-H 'Content-Type: application/json' \
-d '{"message":"Hi","user_id":"u1","session_id":"s1","stream":false}'
Interactive API docs at http://localhost:8000/docs. Full quickstart guide and API reference →
Or deploy to the cloud with one click
Who is this for?
AI engineers
Building production agent systems and tired of stitching FastAPI + Postgres + Qdrant + OAuth boilerplate for every project.
Startups
Shipping agent APIs and want a permissive, MIT-licensed runtime they own — not a managed service with usage-based pricing.
Platform teams
Standing up an internal agent platform with multi-tenant isolation, approvals, and observability across many agent use cases.
Agno users
Already building on Agno and want the missing production layer — sessions, tokens, prompts, queues — without re-inventing it.
Deploy options
| Path | Best for | Config |
|---|---|---|
| Render | Easiest first deploy | render.yaml (image-based) |
| Railway | Best Postgres UX | railway.toml |
| Koyeb | Free nano tier | koyeb.yaml |
| Kubernetes | Self-hosting / scale | Kustomize manifests |
| AWS / Azure / GCP | Cloud-native | deploy/ (ECS, Container Apps, Cloud Run) |
Full per-platform playbook with smoke tests and teardown: Deploy guide →
Get started
MIT licensed. Contributions welcome — see issues on GitHub.
FAQ
Is it production-ready?
Yes, with caveats. The gateway, supervisor, queue, and toolkits are covered by an integration test suite that runs against real Postgres + Qdrant on every push. Pick your own observability backend and harden the auth before shipping to end users.
How does it relate to Agno?
Agents Gateway is a service layer on top of Agno. Agno provides the agent primitives — LLM calls, tools, sessions. The gateway provides the HTTP API, multi-tenancy, persistence, supervisor orchestration, OAuth tokens, and operational surface.
Is it a request-routing API gateway like Kong?
No. Despite the name, Agents Gateway is an agent runtime, not an HTTP request router. It serves and orchestrates AI agents over a REST API; it doesn’t proxy traffic between services.
What models does it support?
Anything Agno supports — OpenAI, Anthropic (Claude), Gemini, plus other providers via Agno. Configure per agent via the model field.
Does it need a vector database?
Qdrant is bundled via docker-compose for the knowledge base. If you’re not using knowledge entries, you can run without it. The chat path doesn’t depend on Qdrant.
Can I run it without Docker?
Yes — you need Postgres 14+ and (optionally) Qdrant reachable over the network. See the deploy guide.