REST endpoints

Every route registered by Server.GenerateRoutes in server/routes.go, grouped by family.

Liveness and version

Method	Path	Handler	Purpose
GET/HEAD	/	inline	"Ollama is running".
GET/HEAD	/api/version	inline	{"version": "..."} from version/.
GET	/api/status	Server.StatusHandler	High-level daemon status.

Native API

Method	Path	Handler	Purpose
POST	/api/pull	Server.PullHandler	Pull a model from the registry. NDJSON progress.
POST	/api/push	Server.PushHandler	Push a model.
GET/HEAD	/api/tags	Server.ListHandler	List local models.
POST	/api/show	Server.ShowHandler	Show model metadata, template, capabilities.
DELETE	/api/delete	Server.DeleteHandler	Delete a local model.
POST	/api/me	Server.WhoamiHandler	Cloud "who am I".
POST	/api/signout	Server.SignoutHandler	Remove the local key from the cloud account.
DELETE	/api/user/keys/:encodedKey	Server.SignoutHandler	Same, addressed by encoded key.
POST	/api/create	Server.CreateHandler	Create a model from a Modelfile.
POST	/api/blobs/:digest	Server.CreateBlobHandler	Upload a blob (used during create).
HEAD	/api/blobs/:digest	Server.HeadBlobHandler	Check if a blob exists locally.
POST	/api/copy	Server.CopyHandler	Copy a model under a new name.
GET	/api/ps	Server.PsHandler	Loaded runners.
POST	/api/generate	Server.GenerateHandler	Single-prompt completion (NDJSON). Wrapped by request logging.
POST	/api/chat	Server.ChatHandler	Multi-message chat (NDJSON). Wrapped by request logging.
POST	/api/embed	Server.EmbedHandler	Vector embeddings.
POST	/api/embeddings	Server.EmbeddingsHandler	Legacy embeddings shape.

Experimental

Method	Path	Handler	Purpose
POST	/api/experimental/web_search	Server.WebSearchExperimentalHandler	Cloud-backed web search.
POST	/api/experimental/web_fetch	Server.WebFetchExperimentalHandler	Cloud-backed URL fetch.
GET	/api/experimental/model-recommendations	Server.ModelRecommendationsExperimentalHandler	Cached recommendations for the launcher.

OpenAI-compatible

Method	Path	Middleware (in order)	Final handler
POST	/v1/chat/completions	withInferenceRequestLogging, cloudPassthroughMiddleware, middleware.ChatMiddleware	ChatHandler
POST	/v1/completions	withInferenceRequestLogging, cloudPassthroughMiddleware, middleware.CompletionsMiddleware	GenerateHandler
POST	/v1/embeddings	cloudPassthroughMiddleware, middleware.EmbeddingsMiddleware	EmbedHandler
GET	/v1/models	middleware.ListMiddleware	ListHandler
GET	/v1/models/:model	cloudModelPathPassthroughMiddleware, middleware.RetrieveMiddleware	ShowHandler
POST	/v1/responses	withInferenceRequestLogging, cloudPassthroughMiddleware, middleware.ResponsesMiddleware	ChatHandler
POST	/v1/images/generations	cloudPassthroughMiddleware, middleware.ImageGenerationsMiddleware	GenerateHandler
POST	/v1/images/edits	cloudPassthroughMiddleware, middleware.ImageEditsMiddleware	GenerateHandler
POST	/v1/audio/transcriptions	middleware.TranscriptionMiddleware	ChatHandler

Anthropic-compatible

Method	Path	Middleware	Handler
POST	/v1/messages	withInferenceRequestLogging, cloudPassthroughMiddleware, middleware.AnthropicMessagesMiddleware	ChatHandler

Behavior notes

• Inference handlers wrapped with withInferenceRequestLogging are captured to the on-disk request log (configured via OLLAMA_INFERENCE_LOG_DIR).
• Cloud passthrough runs before compat middleware. Cloud-only models bypass local handlers entirely with their original request body.
• Streaming endpoints emit NDJSON for native routes (/api/...) and SSE for OpenAI-compatible routes when stream: true is set.
• Authorization is signed by the local SSH key for any request the daemon forwards to the cloud (see auth).

Reference docs

Source	Where
User-facing API reference	docs/api.md
OpenAPI spec	docs/openapi.yaml
Go client types	api/types.go