Local API

Use the bdebd daemon from other tools. OpenAI-compatible chat and completion endpoints, fill-in-the-middle, and management endpoints.

The daemon listens on http://127.0.0.1:41337 (loopback only). Anything under /v1/ plus llama.cpp’s native endpoints are proxied to the engine; the engine is started on demand if it is idle.

Authentication: the API token

Every request needs the per-install token. It is a random 64-character string created on first run and stored in the daemon data directory as a file named token:

OS	Path
Windows	`%LOCALAPPDATA%\BdebTech\AI\token`
macOS	`~/Library/Application Support/BdebTech/AI/token`
Linux	`~/.local/share/bdebtech/ai/token`

Print it with bdebd token. Send it as X-Bdeb-Token: <token> or, for OpenAI-style clients, as the API key (Authorization: Bearer <token>). Requests without it get 401.

Why: the daemon is loopback-only, but any web page you visit can still fire requests at 127.0.0.1:41337. A browser cannot read the token file, so the token keeps web pages out while every local tool on your machine can read it. Browser-based CORS is granted only to editor webviews. If you really want an open daemon (for example inside a locked-down VM), start it with bdebd serve --no-token or set "requireToken": false in config.json.

OpenAI-compatible

TOKEN=$(bdebd token)
curl http://127.0.0.1:41337/v1/chat/completions \
  -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"messages":[{"role":"user","content":"Write a Python fizzbuzz"}],"max_tokens":200,"stream":false}'

Supported: /v1/chat/completions (streaming and non-streaming, including response_format with json_schema), /v1/completions, /v1/models, /v1/embeddings (when an embedding model is loaded). The model field is ignored; whatever model is active is used.

Pointing other tools at it

Use the token as the API key everywhere.

Tool	Setting
Cline / Roo Code	Provider: OpenAI Compatible, Base URL `http://127.0.0.1:41337/v1`, API key = your token
Continue	`"provider": "openai", "apiBase": "http://127.0.0.1:41337/v1", "apiKey": "<token>"`
Aider	`aider --openai-api-base http://127.0.0.1:41337/v1 --openai-api-key $(bdebd token) --model openai/local`
Open WebUI	OpenAI API connection with the same base URL and the token as key

Fill-in-the-middle (autocomplete)

curl http://127.0.0.1:41337/infill -H "X-Bdeb-Token: $TOKEN" -H "Content-Type: application/json" -d '{
  "input_prefix": "def add(a, b):\n    return ",
  "input_suffix": "\n\nprint(add(2, 3))\n",
  "n_predict": 32, "temperature": 0.1, "stop": ["\n"]
}'

Returns { "content": "a + b", ... }. This is llama.cpp’s native /infill; the model’s FIM tokens are inserted automatically.

Management endpoints

Method and path	Purpose
`GET /bdeb/status`	Version, hardware profile, engine state, installed models, downloads, license
`GET /bdeb/hw`	Hardware profile only
`GET /bdeb/models`	Registry with `installed`, `recommended`, `active`, `allowed` flags
`POST /bdeb/models/pull` `{ "modelId" }`	Start a download (202). Progress via events
`POST /bdeb/models/cancel` `{ "modelId" }`	Cancel a download
`POST /bdeb/models/use` `{ "modelId" }`	Switch the served model (downloads if needed, restarts engine)
`POST /bdeb/models/remove` `{ "modelId" }`	Delete model files
`POST /bdeb/engine/start` · `/stop` · `/restart`	Engine lifecycle
`GET /bdeb/config` · `POST /bdeb/config`	Read or patch daemon config
`GET /bdeb/license` · `POST /bdeb/license` `{ "key" }` · `DELETE /bdeb/license`	License management
`GET /bdeb/events`	Server-sent events: `status`, `download`, `engine`, `index`, `log` (token also accepted as `?token=` here, for EventSource)
`GET /bdeb/ping`	Liveness without a token: `{ ok, auth }`
`POST /bdeb/shutdown`	Exit the daemon

CLI

bdebd serve [--port N] [--verbose] [--no-token]
bdebd setup [--model <id>] [--no-verify]
bdebd status | hw | models | token
bdebd models add <path|url> [--id X] [--name N] [--ctx 8192]
bdebd models remove <model-id>
bdebd pull <model-id> | use <model-id>
bdebd license <key> | license --remove
bdebd stop | version

Model ids

qwen2.5-coder-0.5b-q8 · qwen2.5-coder-1.5b-q4 · qwen2.5-coder-7b-q4 · qwen2.5-coder-14b-q4 · qwen2.5-coder-32b-q4