Get Rind to use the model you specify

Goal: connect a provider, select a model, pin defaults per project, and know where to check the current state.

First connection

/login

The menu covers three kinds of sources:

Source Best for Notes
Provider list + API Key OpenAI, Anthropic, Google, DeepSeek, etc. Keys are kept in the local credentials store and never written into the repo
ChatGPT subscription sign-in You already have a ChatGPT subscription Browser authorization; reuses the subscription quota
Named endpoint Self-hosted or third-party OpenAI-compatible services Give it a name, then select by name later; keys and addresses are managed together

/logout [provider] removes a stored credential.

Choosing a model and reasoning effort

  • /model (bare command) opens the selection menu; /model set <model> sets it directly.
  • /effort shows or sets the reasoning effort: low | medium | high | xhigh | max; Ctrl+T cycles it within a session.
  • /status shows it all on one screen: session ID, name, the settings file in use, the connection and its endpoint/model, the reasoning effort with where each value came from, and the latest sampling usage.

Model capabilities (context window, whether images are supported) are decided by the built-in catalog, falling back to a cache when a remote refresh fails. When a model explicitly does not support images, the session warns instead of forcing them through; when capabilities are unknown, it warns and then tries, but never records unknown as supported.

Let the project remember defaults

A model chosen in a session only affects that session. To make new sessions in a directory default to a certain model:

rind config set model <model> --folder [dir] [--connection <id>]
rind config set effort <level> --folder [dir]
rind config unset model|effort --folder [dir]

Priority: session > the folder itself (RIND_HOME/workspaces.json, keyed by folder) > user defaults (~/.rind/settings.json). When a Git worktree has no settings of its own, it inherits the values of its main repo.

How images get into the context

Just write the path relative to the workspace uploads/ directory in the prompt, for example:

uploads/screenshot.png what does this error mean

The system snapshots the image into the session (for format, size, and count limits, see Images into context); later requests read from the snapshot and no longer open the original file.

Back to the scenario index