Skip to content
TyrnBeta

Autocomplete

Autocomplete suggests code as you type, using a model from Tyrn router. It is off until you turn it on and choose a model. Open Settings > Agents > Autocomplete.

The Autocomplete settings page with the Enabled switch, Mode and model choices.

Turn it on

  1. Connect a model first. See Connect a model.
  2. Open Settings > Agents > Autocomplete.
  3. Under Models, pick the model for completions.
  4. Turn on Enabled.

The setting behind this is edit_predictions.provider, which is tyrn when on and none (the default) when off.

Accept a suggestion

Press Tab to accept. Alt-Tab also accepts.

Mode

Under Autocomplete, choose how suggestions appear.

  • Eager (default): suggestions appear inline.
  • Subtle: suggestions appear while you hold a modifier key.

Models

  • The first model is the router model used for inline completions (model).
  • The optional second model is for multi-line next-edit predictions (next_edit_model). Leave it unset to use the completion model.

Timing and limits

Setting Default Meaning
max_output_tokens 256 Longest suggestion.
debounce_ms 150 Wait after you stop typing.
deadline_ms 2500 Give up on a completion after this long.
next_edit_deadline_ms 5000 Same, for next-edit predictions.

Exclusions

Files matching an exclusion are never sent for autocomplete. Add your own with Add exclusion. By default Tyrn keeps out .env*, *.pem, *.key, *.cert, *.crt, .dev.vars, secrets.yml and Tyrn settings and keymap files.

Settings

settings.json
{
  "edit_predictions": {
    "provider": "tyrn",
    "mode": "eager",
    "tyrn": {
      "model": "provider/model",
      "next_edit_model": null,
      "max_output_tokens": 256,
      "debounce_ms": 150,
      "deadline_ms": 2500,
      "next_edit_deadline_ms": 5000
    }
  }
}

Replace provider/model with a model from a connection you added. Model ids appear exactly as the provider names them.

Next steps