[Discussion] Robyn-based Runtime with Full LangGraph API Parity #7

Open
opened 2026-02-15 20:17:17 -05:00 by yindo · 0 comments
Owner

Originally created by @l4b4r4b4b4 on GitHub (Feb 5, 2026).

Summary

I've built a Robyn-based alternative runtime for the LangGraph tools agent that achieves full API parity with the standard langgraph dev server. Sharing in case there's interest in adding this as an optional runtime or learning from my implementation.

What I Built

A complete HTTP server using Robyn (Rust-powered Python web framework) that implements the full LangGraph API specification:

Features

  • Full LangGraph API Parity - All Tier 1, 2, and 3 endpoints
  • SSE Streaming - Real-time event streaming for runs
  • Store API - Key-value storage with namespace support
  • Crons API - Scheduled recurring runs (APScheduler)
  • A2A Protocol - Agent-to-Agent communication (JSON-RPC 2.0)
  • MCP Protocol - Model Context Protocol server
  • Prometheus Metrics - /metrics endpoint for observability
  • Supabase JWT Auth - Built-in authentication middleware
  • Owner Isolation - Multi-tenant resource scoping

Test Coverage

  • 440 unit tests covering all endpoints, handlers, and edge cases
  • CI/CD with GitHub Actions (lint, test, Docker build)

Why Robyn?

  • Performance: Rust-powered HTTP runtime, lower latency than pure Python
  • Production-ready: Built-in workers, graceful shutdown
  • Lightweight: Smaller memory footprint than FastAPI + Uvicorn

Architecture

robyn_server/
├── app.py              # Main Robyn application
├── routes/             # HTTP route handlers
│   ├── assistants.py
│   ├── threads.py
│   ├── runs.py
│   ├── streams.py
│   ├── store.py
│   ├── crons.py
│   ├── a2a.py
│   └── mcp.py
├── a2a/                # A2A Protocol implementation
├── crons/              # Crons API with APScheduler
├── mcp/                # MCP Protocol implementation
├── storage.py          # In-memory storage layer
├── auth.py             # Supabase JWT middleware
└── tests/              # 440 tests

Example /info Response

{
  "service": "oap-langgraph-tools-agent",
  "runtime": "robyn",
  "version": "0.0.1",
  "capabilities": {
    "streaming": true,
    "store": true,
    "crons": true,
    "a2a": true,
    "mcp": true,
    "metrics": true
  },
  "tiers": {
    "tier1": true,
    "tier2": true,
    "tier3": true
  }
}

My Fork

The implementation is available in my fork: https://github.com/l4b4r4b4b4/oap-langgraph-tools-agent

Key directories:

  • robyn_server/ - Complete Robyn runtime
  • .github/workflows/ - CI/CD pipelines
  • Docker images published to GHCR

Questions for Maintainers

  1. Is there interest in adding an alternative runtime option to this repo?
  2. Would a standalone package (robyn-langgraph-runtime on PyPI) be more appropriate?
  3. Any feedback on the API implementation or areas I might have missed?

Happy to contribute back if this aligns with the project's direction. Also happy to maintain it as a separate package if that's preferred.


This is not a PR request - just sharing what I built in case it's useful to the community.

Originally created by @l4b4r4b4b4 on GitHub (Feb 5, 2026). ## Summary I've built a **Robyn-based alternative runtime** for the LangGraph tools agent that achieves full API parity with the standard `langgraph dev` server. Sharing in case there's interest in adding this as an optional runtime or learning from my implementation. ## What I Built A complete HTTP server using [Robyn](https://github.com/sparckles/robyn) (Rust-powered Python web framework) that implements the full LangGraph API specification: ### Features - ✅ **Full LangGraph API Parity** - All Tier 1, 2, and 3 endpoints - ✅ **SSE Streaming** - Real-time event streaming for runs - ✅ **Store API** - Key-value storage with namespace support - ✅ **Crons API** - Scheduled recurring runs (APScheduler) - ✅ **A2A Protocol** - Agent-to-Agent communication (JSON-RPC 2.0) - ✅ **MCP Protocol** - Model Context Protocol server - ✅ **Prometheus Metrics** - `/metrics` endpoint for observability - ✅ **Supabase JWT Auth** - Built-in authentication middleware - ✅ **Owner Isolation** - Multi-tenant resource scoping ### Test Coverage - **440 unit tests** covering all endpoints, handlers, and edge cases - CI/CD with GitHub Actions (lint, test, Docker build) ### Why Robyn? - **Performance**: Rust-powered HTTP runtime, lower latency than pure Python - **Production-ready**: Built-in workers, graceful shutdown - **Lightweight**: Smaller memory footprint than FastAPI + Uvicorn ## Architecture ``` robyn_server/ ├── app.py # Main Robyn application ├── routes/ # HTTP route handlers │ ├── assistants.py │ ├── threads.py │ ├── runs.py │ ├── streams.py │ ├── store.py │ ├── crons.py │ ├── a2a.py │ └── mcp.py ├── a2a/ # A2A Protocol implementation ├── crons/ # Crons API with APScheduler ├── mcp/ # MCP Protocol implementation ├── storage.py # In-memory storage layer ├── auth.py # Supabase JWT middleware └── tests/ # 440 tests ``` ## Example `/info` Response ```json { "service": "oap-langgraph-tools-agent", "runtime": "robyn", "version": "0.0.1", "capabilities": { "streaming": true, "store": true, "crons": true, "a2a": true, "mcp": true, "metrics": true }, "tiers": { "tier1": true, "tier2": true, "tier3": true } } ``` ## My Fork The implementation is available in my fork: https://github.com/l4b4r4b4b4/oap-langgraph-tools-agent Key directories: - `robyn_server/` - Complete Robyn runtime - `.github/workflows/` - CI/CD pipelines - Docker images published to GHCR ## Questions for Maintainers 1. **Is there interest** in adding an alternative runtime option to this repo? 2. **Would a standalone package** (`robyn-langgraph-runtime` on PyPI) be more appropriate? 3. **Any feedback** on the API implementation or areas I might have missed? Happy to contribute back if this aligns with the project's direction. Also happy to maintain it as a separate package if that's preferred. --- *This is not a PR request - just sharing what I built in case it's useful to the community.*
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: langchain-ai/oap-langgraph-tools-agent#7