Workspace and tools

Search and usage

Method	Path	Returns
GET	/search?q=invoice&limit=60	SearchHit[]: messages in every chat containing q, newest first, with an excerpt and where the match is
GET	/usage?days=30	UsageRow[]: tokens, requests and cost by day, bot and model (1–366 days)

Export and import

Method	Path	Body	Returns
POST	/export	{ "botIds": null, "chats": true, "settings": true }	{ "data": "<base64 .inkbot file>" }
POST	/import	{ "data": "<base64>", "bots": true, "chats": true, "settings": true, "preview": true }	ImportResult: what was (or with preview, would be) added and updated, and what's left to do

botIds: null exports every bot. Exports never contain passwords, keys or sign-ins.

Skills

Method	Path	Body	Returns
GET	/skills		SkillInfo[]
GET	/skills/{name}		SkillDetail: the SKILL.md content, its files and requirements, or 404
GET	/skills/{name}/file?path=scripts/run.py		{ path, base64 }, or 404
PUT	/skills	{ "name", "content", "enabled", "source", "bots", "fileChanges" }	SkillDetail. content is the whole SKILL.md; bots sets which bots have it ticked.
DELETE	/skills/{name}		204
POST	/skills/fetch	{ "url": "https://github.com/…" }	The skill's files from a link, for review. Nothing is saved.
GET	/skill-drafts		SkillDraft[]: skills bots drafted from their work
POST	/skill-drafts/resolve	{ "id", "approve": true, "scope": "team" }	SkillDraft. scope is bot or team; send name, description or body to approve an edited version.

In fileChanges, each { path, base64 } adds or replaces a file, and base64: null removes it.

Team memory

Method	Path	Body	Returns
GET	/team-memory		BotMemory
PUT	/team-memory	MemoryEntry[]	BotMemory. Replaces every fact.

Background tasks and terminals

Method	Path	Body	Returns
GET	/background-tasks?botId=scout		BackgroundTaskInfo[], newest first (all bots without botId)
GET	/background-tasks/{id}/output?since=0		{ task, text, next, trimmed }. Pass next as since to read only new output.
POST	/background-tasks/{id}/stop		204
DELETE	/background-tasks/{id}		204. Removes a finished task from the list.
GET	/terminals/shells		ShellProfile[]: the shells this server can open
POST	/terminals	{ "shell": "bash", "cols": 120, "rows": 30 }	BackgroundTaskInfo for the new shell. Read its output like a task's.
POST	/terminals/{id}/input	{ "text": "ls -la\r" }	204. Keys typed in, including control characters (\u0003 is Ctrl+C).
POST	/terminals/{id}/size	{ "cols": 160, "rows": 40 }	204
DELETE	/terminals/{id}		204. Ends the shell and everything running in it.

The terminal endpoints also work with a bot's task id, to type into it.

Computers

Method	Path	Returns
GET	/computers	{ dockerReady, dockerProblem, imageReady, computers, freeMemoryMb, totalMemoryMb }
GET	/computers/{botId}	{ botId, state, note, userHasControl, handoverReason }; state is None, Off, Starting, Waiting or Running
POST	/computers/{botId}/open	{ "url": "…" }: starts it if needed and returns a link to watch it (valid for 12 hours)
POST	/computers/{botId}/stop	204
POST	/computers/{botId}/reset	204. Deletes its apps, browser data and sign-ins.
POST	/computers/{botId}/take	204. You take control, and the bot waits.
POST	/computers/{botId}/handback	204

Local models

Method	Path	Body	Returns
GET	/local-models?updates=true		The runtime, GPU, downloaded models and downloads in progress
POST	/local-models/runtime/install		204. Installs or updates llama.cpp in the background.
GET	/local-models/hub?q=qwen&sort=best		GGUF models on Hugging Face. sort: best, trending, downloads, likes, updated, newest.
GET	/local-models/hub/repo?id=owner/name		A model's quants, each with whether it fits
POST	/local-models/downloads	{ "repo", "quant", "files" }	The download's state. It carries on in the background.
POST	/local-models/downloads/cancel	{ "id": "<download id>" }	204
POST	/local-models/model	{ "id", "settings", "load": true }	204. Saves settings and/or loads (true) or unloads (false).
POST	/local-models/preview	{ "id", "settings" }	Where the model's memory would go with those settings
POST	/local-models/model/delete	{ "id" }	204
GET	/local-models/log?id=…		{ "text": "…" }: what llama-server printed
PUT	/local-models/options	{ "idleMinutes", "useLmStudioModels", "useAllGpus", "huggingFaceToken", "folder" }	204. Fields left out stay as they are.

Model ids contain slashes, so they go in the body rather than the path.