| name | kiln-cli |
| description | Use Kiln as a pure OpenAI-compatible CLI for local chat, translation, embeddings, OCR, model listing, and backend health checks. Trigger whenever the user asks to use a local model, translate text, generate embeddings, OCR an image, inspect available models, or diagnose the local API. |
| compatibility | Requires the `kiln` executable and an OpenAI-compatible backend reachable through `KILN_BASE_URL`. |
Kiln CLI
Kiln is only an API client. It does not start, stop, restart, unload, configure, or supervise the backend. Do not use launchd commands, kiln service, kiln serve, kiln unload, or kiln config; those are no longer CLI commands.
Keep the API key private. Never print it, copy it into an artifact, or put it in a shell command shown to the user. The local default key is local; use KILN_API_KEY or KILN_API_KEY_FILE when the backend expects another value.
Environment
export KILN_BASE_URL=http://127.0.0.1:8007
export KILN_API_KEY=local
export KILN_MODEL=ornith-ai--Ornith-1.5-35B-A3B-MLX-4bit
export KILN_OCR_MODEL=Unlimited-OCR-mxfp8
export KILN_EMBEDDING_MODEL=mlx-community--Qwen3-Embedding-4B-4bit-DWQ
export KILN_TRANSLATE_MODEL=Hy-MT2-1.8B-4bit
OCR and translation fall back to KILN_MODEL when their specific model variable is unset. Embeddings have no fallback and fail clearly when is unset.
