OpenAI-compatible, down to the wire format
Bee speaks the Chat Completions protocol your code already uses. Switch the base URL and the API key — streaming, tool calling, JSON mode, and vision keep working.
Migration is a config change, not a rewrite
Anywhere you can set an OpenAI base URL, you can point at Bee: the official OpenAI SDKs, LangChain, LiteLLM, or a plain HTTP client. Set the base URL to https://bee.heossi.com/bee, use a Bee API key, and pick a Bee model alias such as bee-comb.
The request and response shapes are the ones your code already parses — messages, choices, usage, streamed server-sent events.
What the API supports
- →Streaming responses on every tier.
- →Tool / function calling with structured tool_calls on Bee Comb, Buzz, and Hive.
- →JSON mode with schema-guided structured output on Comb, Buzz, and Hive.
- →Image and video input on Comb, Buzz, and Hive — send media the same way you would to any multimodal Chat Completions endpoint.
- →Document retrieval (RAG): pass document references and Bee grounds answers with citations.
Works in the editors you already use
Because the protocol is standard, editor integrations need no plugin work: Cursor, Windsurf, Continue.dev, Zed, JetBrains AI Assistant, and Aider all accept an OpenAI-compatible base URL and key. VS Code additionally has a native Bee extension with inline chat.
There is also a hosted MCP server — one uvx command connects Bee to Claude Desktop, Claude Code, Cursor, VS Code, Zed, Windsurf, or OpenCode as a set of callable tools.
A model ladder instead of a model lottery
Model aliases map to tiers with published per-token rates — from Bee Cell at $0.15 per 1M input / $0.60 per 1M output up to Bee Swarm at $5 / $15. Paid tiers run a 256K-token native context window; Comb and Buzz add a 1M-token long-context mode, and Swarm serves 1M natively.
Every alias resolves to a dated, immutable snapshot, so you can reproduce a deployed model state instead of chasing a floating label.
Frequently asked questions
- Do I need a new SDK to use Bee?
- No. Any OpenAI-compatible client works — set the base URL to https://bee.heossi.com/bee and pass a Bee API key. Bee also publishes its own SDKs and a hosted MCP server if you prefer them.
- Does tool calling return the same shape as OpenAI's API?
- Yes — assistant messages carry structured tool_calls entries that your existing tool-dispatch loop can execute unchanged. Tool calling is served on Bee Comb, Buzz, and Hive.
- Is JSON mode supported?
- Yes. Comb, Buzz, and Hive accept response_format and return schema-valid JSON via guided decoding, which is stronger than prompt-level “please answer in JSON”.
- Can I send images or video through the same endpoint?
- Yes — Comb, Buzz, and Hive accept image and video input natively through the standard multimodal message format.
- Is there a free way to test compatibility?
- API Free includes 500K pooled tokens per month with one key and no card, which is enough to run your integration tests against the real endpoint.
Related
Start on the free tier
Bee Cell is free — no card. Scale to paid tiers, the API, or sovereign deployment when you are ready.