AI infrastructure engineers don't have to fight
OpenAI-compatible endpoints in every editor, tool calling for automation, a 1M-token window for monorepo-scale questions, and team billing that finance understands.
Meets your toolchain where it is
Bee speaks the OpenAI Chat Completions wire format, so Cursor, Windsurf, Continue.dev, Zed, JetBrains AI Assistant, and Aider connect by changing a base URL and key. VS Code gets a native extension; every MCP client — Claude Desktop, Claude Code, Cursor, and others — connects to the hosted Bee MCP server with one uvx command.
No proprietary SDK lock-in: the integration surface is the protocol your tools already implement.
Programming is a first-class domain
- →Code generation, explanation, refactoring, and test writing across the tier ladder — programming is one of Bee's active platform domains.
- →Tool calling on Comb, Buzz, and Hive: wire the model into linters, test runners, and internal APIs with structured tool_calls.
- →JSON mode for machine-consumable output in CI — review annotations, migration plans, dependency reports.
- →1M-token long-context on Comb and Buzz: cross-cutting questions over a large source tree in one request.
- →Vision and video input: paste a failing UI screenshot or a screen recording of a bug straight into the conversation.
Team mechanics, solved once
Engineering teams start at Bee Comb ($79/mo, 3 seats, 10M pooled tokens) and scale to Buzz for heavier agent workloads or Hive for the 99.5% uptime SLA. Pooled tokens mean one budget across the team; per-seat additions are a published rate on one Stripe invoice, not a procurement cycle.
The free Cell tier gives every engineer a real endpoint (128K context) for evaluation, and API Free covers integration tests without a card.
Answers you can reproduce
Every model alias resolves to a dated immutable snapshot, so a pipeline pinned today produces attributable behaviour next quarter. When something changes, the release is documented in the public changelog with validation evidence at /trust — the same discipline you'd expect from any dependency you take into a build.
Customer data is not used to train the Bee base model. Prompts, documents, and outputs stay yours; processing is limited to serving, securing, and operating the service.
Frequently asked questions
- Which editors work with Bee?
- Anything that accepts an OpenAI-compatible base URL: Cursor, Windsurf, Continue.dev, Zed, JetBrains AI Assistant, Aider, and more. VS Code also has a native Bee extension, and MCP clients connect via the hosted Bee MCP server.
- Can we use Bee inside CI pipelines?
- Yes — the API is standard Chat Completions with streaming, tool calling, and JSON mode, so CI jobs call it like any HTTP service and parse schema-valid output.
- How does Bee handle very large codebases?
- Comb and Buzz serve a dedicated 1M-token long-context mode for repository-scale requests, and document retrieval covers corpora beyond any window with cited answers.
- Is our source code used for training?
- No. Customer data — including code sent through the API or workspace — is not used to train the Bee base model.
- How do we trial it with the whole team?
- Every engineer can start free on Cell today; when you want shared billing and pooled tokens, Comb starts at $79/mo with 3 seats included.
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.