Built an Agent Protocol server with FastAPI - open-source LangGraph Platform alternative

Posted by Lost-Trust7654@reddit | Python | View on Reddit | 6 comments

Hey Python community!

I've been building an Agent Protocol server using FastAPI and PostgreSQL as an open-source alternative to LangGraph Platform.

What My Project Does:

Serves LangGraph agents via HTTP APIs following the Agent Protocol specification
Provides persistent storage for agent conversations and state
Handles authentication, streaming responses, and background task processing
Offers a self-hosted deployment solution for AI agents

Target Audience:

Production-ready for teams deploying AI agents at scale
Developers who want control over their agent infrastructure
Teams looking to avoid vendor lock-in and expensive SaaS pricing
LangGraph users who need custom authentication and database control

Comparison with Existing Alternatives:

LangGraph Platform (SaaS): Expensive pricing ($500+/month), vendor lock-in, no custom auth, forced tracing
LangGraph Platform (Self-hosted Lite): No custom authentication, limited features
LangServe: Being deprecated, no longer recommended for new projects
My Solution: Open-source, self-hosted, custom auth support, PostgreSQL persistence, zero vendor lock-in

Agent Protocol Server: https://github.com/ibbybuilds/agent-protocol-server

Tech stack:

FastAPI for the HTTP layer
PostgreSQL for persistence
LangGraph for agent execution
Agent Protocol compliance

Status: MVP ready, working on production hardening. Looking for contributors and early adopters.

Would love to hear from anyone working with LangGraph or agent deployment!

[-]

We have been building something similar at RunAgent too. Still early, but working hard to have a standardised agent deployment server accross agentic platforms, and accessible through language specific SDKs.

https://github.com/runagent-dev/runagent

Lost-Trust7654@reddit (OP)

I read the documentation and found out how to invoke and stream a graph from LangGraph, but what about persistence and authentication?

Thanks! You guys are doing an amazing job, and starred as well.

snailspeed25@reddit

This is super interesting, especially since I've never gotten to work with LangGraph before. It's a much better alternative to current solutions that I've seen so far. Feel free to post this in this skool community for more people to see it since this could be super useful https://www.skool.com/python-2696/about?ref=9d926b02de79408487d6dde1905724a0

Thanks. I will post there as well.