Framework Integrations
Ferro Labs AI Gateway exposes an OpenAI-compatible HTTP surface, so any LLM framework that talks to OpenAI works against Ferro by setting base_url. The pages below go further — they show first-party adapters, framework-idiomatic code, and the one piece of glue worth knowing: trace_id.
The trace_id join key
Every Ferro response (since ai-gateway v1.1.0) carries a trace_id in the x-trace-id HTTP header. First-party adapters surface it on response_metadata["trace_id"] (Python) and response.traceId (TypeScript). That single ID is the join key between your framework's instrumentation and any observability backend wired into the gateway via a plugin bridge — LangSmith, Langfuse, Phoenix, Datadog, and more.
You write framework code once. Telemetry routing changes by swapping a gateway plugin, not by editing your app.
Python
| Framework | Status | Page |
|---|---|---|
| LangChain (Python) | ✅ langchain-ferrolabsai 0.1.0 on PyPI | LangChain Python |
| LangGraph | ✅ Works today via FerroChatModel | LangGraph |
| LangSmith | ✅ Trace correlation today · 🚧 dedicated bridge plugin in v1.2 | LangSmith |
| LlamaIndex | 🚧 llama-index-llms-ferrolabsai 0.0.1 placeholder on PyPI | LlamaIndex |
| CrewAI | 🚧 Drop-in via OpenAI-compatible client | CrewAI |
| AutoGen | 🚧 Drop-in via OpenAI-compatible client | AutoGen |
| Haystack | 🚧 Drop-in via OpenAI-compatible client | Haystack |
| DSPy | 🚧 Drop-in via OpenAI-compatible client | DSPy |
| Pydantic AI | 🚧 Drop-in via OpenAI-compatible client | Pydantic AI |
| Instructor | 🚧 Drop-in via OpenAI-compatible client | Instructor |
TypeScript / JavaScript
| Framework | Status | Page |
|---|---|---|
| Vercel AI SDK | ✅ Drop-in via @ai-sdk/openai or @ferro-labs-ai/sdk | Vercel AI SDK |
| LangChain.js | 🚧 Dedicated sub-export @ferro-labs-ai/sdk/langchain planned | LangChain.js |
| Mastra | 🚧 Drop-in via Vercel AI SDK provider | Mastra |
Runnable examples
Every page links to a Dockerized recipe in the ai-gateway-cookbook repo. From a fresh clone:
cd python/02-langgraph-multi-provider-agent
cp .env.example .env # fill in FERRO_API_KEY
make run # one command, working multi-provider agent
That recipe — planner on GPT-4o, coder on Claude, summarizer on Gemini, all through one gateway URL — is the smallest demo that proves Ferro's value inside a framework you already use.