Docs · Guide
Snow CLI on any model
Snow CLI (snow, the npm package snow-ai) keeps each API setup as a profile. magpie adds one of its own that points Snow at magpie’s gateway, so Snow can run on any model you have in magpie: DeepSeek, GLM, Qwen, an OpenAI or Anthropic key, or a ChatGPT or Copilot subscription.
Snow keeps its settings in ~/.snow (or $SNOW_CONFIG_DIR). Each profile is a file in profiles/, active-profile.json names the one in use, and config.json is a copy of it. magpie writes profiles/magpie.json: a copy of the profile you were on, with magpie’s gateway as its base URL, a key that names Snow, Chat Completions as its protocol and the model you pick as its model. It makes that profile the active one and copies it to config.json, as Snow’s own profile switch does. The gateway then sends each request to the model’s provider.
Connect Snow
- Add providers in magpieIn the Providers tab, add an API key or sign in to a subscription.
- Pick a model on Snow CLI’s rowIn the Agents tab, choose a model under magpie.
- Restart open snow sessionsSnow reads its profile at start-up.
magpie snow deepseek/deepseek-v4-pro # Snow on DeepSeek magpie snow default # back to your own profile
It works the same for Snow in a WSL distro on Windows: the profile there names the gateway as the distro reaches it.
Your profiles
Your own profiles are not changed. magpie’s copies your other settings from the one you were on, but leaves out a subscription sign-in of yours (snowcfg.oauth). A profile of your own that happens to be named magpie is kept and put back. On the row, the models of your own profiles are listed too: picking one makes Snow switch to that profile.
Windows and images
magpie sets maxContextTokens, maxTokens and supportsVision from its catalog for the model you pick, so Snow knows its real window and whether it takes images. A model magpie knows no limits for gets Snow’s own defaults, 200K and 64K. If you change one of these in Snow’s settings, it stays as you set it. A vision model of your own is kept only with an endpoint and key of its own. Images for a model that can’t see go through magpie’s Vision, which describes them at the gateway.
The light model
Snow asks a light model (basicModel) for summaries, compaction and file search. On magpie’s profile it follows the main model, unless you pick another magpie model in the row’s small field. That pick stays when you change the main model.
Thinking
The row’s thinking field sets chatThinking on magpie’s profile: a level sends reasoning_effort with each request, and off is Snow’s default. Only models that list reasoning levels offer it.
Snow App
Snow App keeps API setups of its own. Its Sync Snow CLI API config button copies Snow CLI’s profiles into it, magpie’s among them, so the app can use magpie’s models too.
Switch back
Pick one of your own models on the row, run magpie snow default, or take the row off magpie. Snow goes back to the profile you were on, profiles/magpie.json is removed, and config.json is a copy of your profile again.
FAQ
Can Snow CLI use DeepSeek, GLM or Qwen?
Yes. Any model you have in magpie can be Snow’s model: DeepSeek, GLM, Qwen, GPT, Claude and more.
Does magpie change my own Snow profiles?
No. It adds a profile of its own, profiles/magpie.json, and makes it the active one. Switching back makes your profile active again.
Why does Snow show the wrong context size?
magpie sets maxContextTokens and maxTokens from its catalog for the model you pick. A value you changed in Snow yourself stays as you set it.
Which protocol does Snow use with magpie?
Chat Completions (requestMethod chat). The row warns when magpie’s profile is set to another one.
More agents on the same models: Models in Gemini CLI, Cursor Private Inference on any model and Antigravity on any model. Something fails? See Fixing common errors, or ask on Discord.