Ollama Compatibility

Deprecated. The Ollama compat layer is kept so existing tools and scripts that hardcoded the Ollama API can keep working while you migrate them. For new clients, use the OpenAI-compatible endpoints under /v1/* — they're better supported, get the most features, and are the focus of ongoing work.

If you have a tool that already calls Ollama's HTTP API (the CLI, a homegrown script, an old IDE plugin, etc.), point its base URL at Bodhi and it should work for the supported endpoints below.

Supported endpoints

Bodhi implements the three Ollama endpoints that real-world clients depend on most:

Endpoint	Method	Purpose
`/api/tags`	GET	List available models (Bodhi's combined catalog reshaped into Ollama's tag format)
`/api/show`	POST	Show model details for one tag
`/api/chat`	POST	Chat with a model (streaming and non-streaming)

That's the entire supported surface. Anything not listed here isn't supported — /api/generate, /api/pull, /api/push, /api/create, /api/copy, /api/delete, /api/embeddings (the Ollama-shaped one), the model-management endpoints, and so on. Don't enumerate further; assume the rest is absent.

Auth

Authorization: Bearer <bodhi-api-token>

Same Bodhi-issued token as every other endpoint on the server. The Ollama CLI doesn't natively send a Bearer token — see its docs for the env var (OLLAMA_HOST plus a custom auth header config) or use a wrapper.

Quick check

curl http://localhost:1135/api/tags \
  -H "Authorization: Bearer $BODHI_TOKEN"

The models array in the response lists the same things GET /v1/models returns, just in Ollama's tag format.

Migration tip

If you're touching the calling code anyway, switch from /api/chat to /v1/chat/completions. The OpenAI format is more widely supported, has richer tool-calling and streaming semantics, and is the path getting active investment. See OpenAI Chat Completions for the equivalent calls.

For embeddings, use /v1/embeddings — Bodhi does not expose an Ollama-format embeddings endpoint.

Full schema