Conifer — run AI locally, route the rest

conifer
An ASCII luminance dither of a conifer, framed by a sun and moon. Hover with a pointer to ripple the characters along the row.

Works where you already work.

Conifer serves OpenAI and Anthropic compatible endpoints on one base URL.

  • Claude Desktop
  • Codex Desktop
  • Antigravity
  • Terminal
  • cmux
  • VS Code
  • OpenCode
  • Pi
  • Zed
  • Cursor

Use our gateway, bring your own key, or bring your own hardware.

  1. Install the runtime

    One command, and the gateway is running on this machine. Point any OpenAI or Anthropic client at it and it answers.

    curl -fsSL https://www.conifer.build/install-cli.sh | sh

  2. Bring your existing keys

    One key covers OpenAI, Anthropic, and OpenRouter. Conifer reads the model name, sends the request to the provider that serves it, and bills land on your account, not ours.

  3. Or run it on your own hardware

    Serve the weights from this computer, from a machine on your LAN, or from a server you run on prem. Queries never leave a network you control, and that path costs nothing per token.

The router reads the question and picks the model.

A question about a port number went to Kimi K3 and came back in three seconds. A question about KV cache limits, with the dial moved to best, went to Claude Opus 5 and took a minute. Same session, nothing restarted.

You choose where every token runs.

Moving a model from your laptop to a server, or from your own key to ours, is a routing change. The base URL does not move, so the code that calls it never has to.

  • your hardwareweights resident on this machine
  • on prema box you own, in your building
  • your LANanother machine on the network
  • your API keysOpenAI, Anthropic, OpenRouter
  • the cloudmanaged, metered per token

one endpoint