Use with OpenAI SDKs & LangChain
Alongside the Anthropic surface, Atlas exposes an OpenAI-compatible API, so OpenAI SDKs, LangChain, llama-index, Continue, Open WebUI, and similar clients work against the same endpoint.
Point an OpenAI client at Atlas
Section titled “Point an OpenAI client at Atlas”from openai import OpenAI
client = OpenAI( base_url="http://localhost:8080/v1", api_key="<your atlas key>",)
resp = client.chat.completions.create( model="qwen3-0.6b", messages=[{"role": "user", "content": "Hello from Atlas"}],)print(resp.choices[0].message.content)What works
Section titled “What works”POST /v1/chat/completions— system prompts, multi-turn, streaming,tools/tool_calls,stop,max_tokens/max_completion_tokens,temperature/top_p,finish_reasonmapping, and usage fields.POST /v1/embeddings— OpenAI embeddings shape, served by anembedding-class model.POST /v1/rerank— the de-facto Cohere rerank shape, served by areranker-class model.GET /v1/models— OpenAI list shape.
Atlas owns this surface rather than proxying the engine’s endpoint, so behavior is identical across engines (vLLM, SGLang, llama.cpp, MLX) and matches the Anthropic surface’s semantics — same auth, aliases, context-window assertion, and error mapping.
See API compatibility for the full surface.