Every day

Models and providers

Lampson talks to models over plain HTTP, in the two formats everyone supports. It has no server of its own: your key is only ever sent to the provider you chose.

provider	format	notes
`deepseek`	OpenAI	inexpensive, strong at code
`anthropic`	Anthropic	Claude models
`openai`	OpenAI	GPT models
`kimi`	OpenAI	Moonshot
`glm`	OpenAI	GLM 5.3 / 5.3 Flash, by Z.ai
`groq`	OpenAI	very fast
`grok`	OpenAI	xAI
`openrouter`	OpenAI	one key, many models
`minimax`	Anthropic
`ollama`	OpenAI	runs on your machine, no key

Switching§

Terminal: /model opens one arrow-key flow — provider → model (listed live from the provider's API) → key, only if it is missing. /model glm glm-5.3-flash applies directly; /model glm asks only for the model. /provider and /setup are the same command.
Web: click the provider · model pill in the header.
The model list comes from the API (GET /models), so you never type a name blind; if the list is long you filter it first, and there is always an "other…" option to type one.

One key per provider is kept, so switching back and forth is instant. Settings live in ~/lampson/.lampson/config.json, on your machine, and are shared by the terminal and the web UI.

Any compatible endpoint§

Set LAMPSON_BASE_URL in ~/lampson/.env (and LAMPSON_WIRE=openai|anthropic if the preset cannot guess):

LAMPSON_PROVIDER=ollama
LAMPSON_BASE_URL=http://127.0.0.1:11434/v1
LAMPSON_MODEL=qwen2.5-coder:14b

Images§

Paste a screenshot into the web composer, or /image file.png in the terminal. If the model you picked does not accept images, Lampson warns you before sending and the model gets a text note instead of an error. Force it either way with "vision": true|false in config.json.

Your key is a secret§

The key is stored as a Synsema secret: the program can send it as a header but can never read, print or log it. .env is ignored by git.

View as Markdown