Skip to content

api

Directory actions

More options

Directory actions

More options

Latest commit

 

History

History

api

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
README.md
README.md
 
 
cli.md
cli.md
 
 
openai-compat.md
openai-compat.md
 
 

README.md

Conifer API

Conifer runs a local HTTP server that speaks two protocols most tools already understand:

  • the OpenAI API (/v1/*), and
  • the Ollama-native API (/api/*).

Point an existing client at the local address and it just works — no account, no key, no data leaving the machine. This is how you build on top of Conifer without touching the engine.

There are two ways to run the server:

  1. The desktop app starts it for you when you enable "local server" (default http://127.0.0.1:8080).
  2. The CLI: conifer serve --model <model.gguf> — see cli.md.

Endpoints

OpenAI-compatible (/v1)

Method Path Purpose
POST /v1/chat/completions Chat completion (streaming via SSE or buffered). The main endpoint.
POST /v1/completions Raw text completion. Supports a suffix for fill-in-the-middle.
POST /v1/embeddings Embedding vectors for one or more inputs.
GET /v1/models List the loaded/available model(s).
POST /infill Fill-in-the-middle (code) completion.

See openai-compat.md for request/response shapes and examples.

Ollama-native (/api)

For tools that auto-discover an Ollama backend (Cline, Roo, Zed, VS Code BYOK):

Method Path Purpose
GET /api/tags List models (Ollama's model list).
POST /api/show Model details + capabilities.
POST /api/chat Chat (NDJSON streaming).
POST /api/generate Raw-prompt generation (NDJSON streaming).
POST /api/embed Embeddings.

Security model

The server binds to loopback (127.0.0.1) by default and has no authentication — it assumes only processes on your machine reach it. It rejects cross-origin browser requests. If you bind a non-loopback interface (--host 0.0.0.0), you are deliberately exposing an unauthenticated engine to your network; do that only behind your own trust boundary.

Compatibility notes

  • The model field in requests is accepted for client compatibility; the server serves whichever model is loaded. (In the desktop app you pick the model in the UI; with the CLI you pass --model.)
  • Streaming chat uses SSE on /v1/chat/completions and NDJSON on the Ollama /api/* endpoints — matching each protocol's native framing.