A Pi coding agent extension that gives the agent the ability to list, search, and switch models on its own.

With this extension, you can tell the agent things like "switch to a cheaper model" or "use Claude for this task" and it will handle the model change itself, without you needing to use /model or keyboard shortcuts.

pi install npm:pi-model-switch

Restart Pi to load the extension.

After restarting Pi, the switch_model tool should be available. Ask the agent to "list available models" to confirm.

If you used curl:

curl -o ~/.pi/agent/extensions/model-switch/index.ts \
  https://raw.githubusercontent.com/nicobailon/pi-model-switch/main/index.ts

If you cloned:

cd ~/.pi/agent/extensions/model-switch && git pull

Restart Pi after updating.

Create aliases.json in the extension directory to define shortcuts:

~/.pi/agent/extensions/model-switch/aliases.json

{
  "cheap": "google/gemini-2.5-flash",
  "fast": "google/gemini-2.5-flash",
  "coding": "anthropic/claude-opus-4-5",
  "budget": ["openai/gpt-5-mini", "google/gemini-2.5-flash", "anthropic/claude-3-5-haiku-latest"]
}

• String value: Must be an available model or returns an error
• Array value: Uses first available model in the list (fallback chain)

Then just say "switch to cheap" or "use coding model".

Add model switching preferences to your AGENTS.md for contextual decisions:

## Model preferences
- Simple file ops / quick questions: switch to "cheap"
- Complex refactoring / architecture: switch to "coding"
- Default to budget-friendly models unless quality is needed

The agent will use aliases when appropriate based on your guidance.

Once installed, the agent gains a switch_model tool. Just ask naturally:

• "List available models"
• "Switch to GPT-5.2"
• "Use Opus 4.5"
• "Change to a model with vision capabilities"
• "Use a cheaper model for this task"

The agent will list models or switch as appropriate.

The extension registers a single tool:

switch_model

Returns all models you have API keys configured for, showing:

• Provider and model ID
• Model name
• Context window and max output tokens
• Capabilities (reasoning, vision)
• Cost per 1M tokens (input/output)
• Which model is currently active

Filters models by partial match on provider, id, or name. Returns all matching models with full details.

Matches models by:

1. Alias lookup (if defined in aliases.json)
2. Exact provider/id match
3. Exact id match
4. Partial match on id, name, or provider

If multiple models match, it asks you to be more specific.

• Pi coding agent
• API keys configured for the models you want to use

MIT