API endpoints reference

Every endpoint KoboldCpp serves, grouped by API family. For request fields, see the API docs at /api on your instance or lite.koboldai.net/koboldcpp_api.

How to read the tables:

• Password: Yes means the endpoint needs the password when one is set (--password). No means it never checks. Admin means admin mode and the admin password. See Passwords and security.
• Needs: the model or flag that must be loaded for the endpoint to do its job. "Text model" is the model in GGUF Text Model:.
• A path in brackets is an alias that does the same.
• A short path after a comma continues the first path's prefix: /api/extra/multiplayer/status, /getstory means /api/extra/multiplayer/getstory.

KoboldAI (native)

Path	Method	Password	Purpose	Needs
/api/v1/generate (/api/latest/generate)	POST	Yes	Generate text	Text model
/api/extra/generate/stream	POST	Yes	Generate text, streamed as server-sent events	Text model
/api/extra/generate/check	GET, POST	Yes	Text generated so far by the running request; POST with genkey for your own request	Text model
/api/extra/abort	POST	Yes	Stop generating; with genkey only your own request	Text model
/api/extra/last_logprobs	GET, POST	Yes	Token probabilities of the last generation	Text model
/api/extra/tokencount (/api/extra/tokenize)	POST	Yes	Count and list tokens; also accepts OpenAI-style messages	Text model
/api/extra/detokenize	POST	Yes	Turn token IDs back into text	Text model
/api/extra/json_to_grammar	POST	Yes	Convert a JSON schema to a grammar
/api/v1/model	GET	Partly	Loaded model name; without a valid key it returns koboldcpp/protected-model
/api/v1/config/max_length, /api/v1/config/max_context_length	GET	No	Limits reported to KoboldAI clients: max_length is 1024 unless --hordegenlen is set (not Default Gen Amt:); max_context_length is the context size, or --hordemaxctx if set
/api/v1/config/soft_prompt, /api/v1/config/soft_prompts_list	GET	No	Compatibility stubs for KoboldAI clients
/api/v1/info/version	GET	No	KoboldAI API version (1.2.5)
/api/extra/version	GET	No	KoboldCpp version and active features
/api/extra/true_max_context_length	GET	No	Context size the server allocated
/api/extra/perf	GET	No	Speed of the last request, queue length
/api/extra/preloadstory	GET	No	Story set with --preloadstory	--preloadstory
/api/extra/websearch	POST	Yes	Web search, body {"q": "..."}; returns [] when off	--websearch
/api/extra/multiplayer/status, /getstory, /setstory	POST	Yes	Shared multiplayer session	--multiplayer
/api/extra/data/list, /load, /save	POST	Yes	Server-side save slots	--savedatafile
/api/extra/transcribe	POST	Yes	Speech-to-text	Whisper model or audio mmproj
/api/extra/tts	POST	Yes	Text-to-speech	TTS model
/api/extra/speakers_list	GET	No	TTS voices
/api/extra/embeddings	POST	Yes	Embeddings	Embeddings model
/api/extra/music/prepare	POST	Yes	Plan a song (audio codes)	Music models
/api/extra/music/generate	POST	Yes	Generate music as WAV or MP3	Music models
/request	POST	Yes	Legacy format: {text, max} in, {"data": {"seqs": [...]}} out	Text model
/api, /docs	GET	No	API documentation

Without --savedatafile, /api/extra/data/list and /load close the connection without an answer, and /save returns HTTP 400.

OpenAI-compatible

Path	Method	Password	Purpose	Needs
/v1/completions (/v1/completion, /completions)	POST	Yes	Text completion	Text model
/v1/chat/completions (/chat/completions)	POST	Yes	Chat, with tools and images; thinking text in reasoning_content	Text model; vision mmproj for images
/v1/responses (/responses)	POST	Yes	OpenAI Responses API	Text model
/v1/models (/models)	GET	No	Loaded model; in router mode also the admin-directory configs
/v1/embeddings	POST	Yes	Embeddings; supports encoding_format: base64	Embeddings model
/v1/audio/transcriptions (/audio/transcriptions)	POST	Yes	Speech-to-text; JSON with audio_data, or a multipart file upload	Whisper model or audio mmproj
/v1/audio/speech (/audio/speech)	POST	Yes	Text-to-speech; response_format: mp3 for MP3	TTS model
/v1/audio/voices (/audio/voices)	GET	No	TTS voices
/v1/images/generations (/images/generations)	POST	No	Generate an image	Image model
/v1/images/edits (/images/edits)	POST	No	Edit an image	Image model
/v1	GET	No	Landing page

Anthropic-compatible

Path	Method	Password	Purpose	Needs
/v1/messages (/messages)	POST	Yes, Bearer only	Anthropic Messages API, with images and tool calls	Text model

Ollama-compatible

Run KoboldCpp with --port 11434 so Ollama apps find it.

Path	Method	Password	Purpose	Needs
/api/generate	POST	Yes	Text completion; streams unless "stream": false	Text model
/api/chat	POST	Yes	Chat with tool calls; streams unless "stream": false	Text model
/api/embed	POST	Yes	Embeddings	Embeddings model
/api/tags, /api/ps	GET	No	Model list
/api/version	GET	No	Reports Ollama version 0.7.0
/api/show	POST	No	Fixed placeholder model info

A1111 / Forge (images)

Path	Method	Password	Purpose	Needs
/sdapi/v1/txt2img	POST	No	Text to image; also video with frames	Image model
/sdapi/v1/img2img	POST	No	Image to image, inpainting	Image model
/sdapi/v1/upscale, /sdapi/v1/extra-single-image	POST	No	Upscale an image	Upscaler (--sdupscaler)
/sdapi/v1/interrogate	POST	No	Describe an image	Vision mmproj; returns 503 No Vision model loaded without
/sdapi/v1/sd-models, /options, /samplers, /schedulers, /loras, /upscalers, /latent-upscale-modes	GET	No	Lists and settings	Image model
/sdapi/v1/progress, /sdapi/v1/get_last.json	GET	No (scoped by genkey)	Progress and last result	Image model

ComfyUI-compatible (images)

Path	Method	Password	Purpose	Needs
/prompt	POST	No	Generate an image; answers when it is done, with a fixed job ID	Image model
/upload/image (/api/upload/image)	POST	No	Upload an input image	Image model
/view (/view.png, /api/view, /view_image)	GET	No	Get the generated image	Image model
/history (/api/history)	GET	No	Emulated history that only holds the last image	Image model
/ping, /system_stats, /object_info	GET	No	Compatibility answers; /system_stats holds placeholder values
/models/checkpoints, /models/loras (also under /api/)	GET	No	Model lists
/ws	GET	No	WebSocket handshake only

XTTS-compatible (speech)

Path	Method	Password	Purpose	Needs
/tts_to_audio	POST	No	Text-to-speech	TTS model
/speakers_list, /speakers	GET	No	Voices
/get_tts_settings	GET	No	Settings stub
/set_tts_settings	POST	No	Accepts anything, changes nothing
/voice/check, /voice/speakers, /voice/vits	GET	No	Legacy TTS endpoints

Admin

Only active with admin mode (Enable Model Administration, --admin). See Admin mode.

Path	Method	Password	Purpose
/api/admin/list_options	GET	Admin	Configs and models you can switch to
/api/admin/reload_config	POST	Admin	Switch config or model, body {"filename": "..."}; unload_model and initial_model are special names
/api/admin/check_state, /save_state, /load_state, /clear_state	POST	Admin	Save and restore the KV cache

Other

Path	Method	Password	Purpose	Needs
/mcp	POST	Yes	MCP JSON-RPC proxy (initialize, tools/list, tools/call)	--mcpfile
/props	GET	No	llama.cpp-style info: chat template, n_ctx, vision and audio support
/slots	GET	No	Always 501; not supported
/.well-known/serviceinfo	GET	No	Service info
/	GET	No	KoboldAI Lite
/noscript	GET	No	Interface without JavaScript
/lcpp/	GET	No	llama.cpp WebUI
/sdui	GET	No	Stable UI for images	Image model to generate
/musicui	GET	No	Music and TTS interface	Music or TTS model to generate
/manifest.json	GET	No	Web app manifest

Not in the official API docs

The API docs at /api leave out: the Ollama, ComfyUI and XTTS endpoints, /v1/images/edits, /v1/audio/voices, /mcp, /request, /sdapi/v1/extra-single-image, the A1111 GET lists other than sd-models, options and samplers, the soft prompt stubs and /slots.