pi-reasoning — pi.dev extension

Automatic reasoning level management for pi.dev. Sets the thinking/reasoning level based on the selected model with sensible defaults for all major models.

No more manually adjusting reasoning every time you switch models — pi-reasoning does it for you.

Features

• 🧠 Auto-reasoning — Model change → optimal thinking level applied instantly
• 📋 Sensible defaults — Covers Claude, GPT, Gemini, DeepSeek, Llama, Mistral, Kimi, o-series
• ⚙️ /reasoning command — Manual override, auto re-apply, or check current level
• 📊 Status bar indicator — Colored dot + level always visible
• 🔌 Zero config — Install and go

Install

pi install npm:@pixu1980/pi-reasoning

Requires Node.js ≥ 22 (uses --experimental-strip-types).

How it works

When you switch models (via /model, Ctrl+P, or session restore), the extension:

1. Checks if the new model supports reasoning
2. Looks up the model in its internal mapping table
3. Applies the corresponding thinking level
4. Updates the status bar with the current level

Non-reasoning models are handled by pi's built-in clamping (always off).

Default mappings

Model family	Thinking level
Claude Opus 4/3, o3, o4, DeepSeek R1	max
Claude Sonnet 4/4.5, GPT-5, Gemini 2.5/3, DeepSeek, Kimi, QwQ	high
Claude 3 Sonnet, GPT-4o/4.1/4, Gemini 2.0 Pro, Llama 3/4, Mistral Large	medium
Claude Haiku/3 Haiku, GPT-4o-mini, Gemini 2.0 Flash, Mistral Small	low
GPT-4.1-mini	minimal
GPT-4.1-nano, GPT-4-mini	off

Models not in the map get a heuristic guess based on their name (nano → off, mini/flash → low, large/pro → high, else medium).

Native pi.dev thinkingLevelMap support

The extension fully respects each model's native thinkingLevelMap as defined by pi.dev. This means:

• Available levels are determined by the model's actual capabilities, not by the extension's mapping table.
• xhigh and max are opt-in levels: they only appear in the menu for models that explicitly declare them in their thinkingLevelMap (e.g. claude-opus-4-7, claude-fable-5).
• Missing thinkingLevelMap: when a model doesn't declare one, only standard levels through high are offered — xhigh/max are hidden.
• Auto-apply clamping: if the extension's recommended level isn't supported by the model, it automatically falls back to the highest available level.

Usage

/reasoning — Show or set reasoning level

/reasoning                    Show current level
/reasoning auto               Re-apply auto-reasoning for current model
/reasoning high               Set specific level (off/minimal/low/medium/high/xhigh/max)
/reasoning reset              Restore default model mappings
/reasoning map                Show active model→level mappings

Examples

# Check current level
/reasoning
# → "Reasoning: 🟡 medium"

# Switch to a reasoning model
/model claude-sonnet-4-20250514
# → Status bar: 🟠 claude-sonnet-...

# Override manually
/reasoning max
# → "Reasoning level → 💜 max"

# Re-apply auto after manual override
/reasoning auto
# → "Auto-reasoning → 🟠 high (anthropic/claude-sonnet-4-20250514)"

# See all mappings
/reasoning map
# → "Active mappings (24):
#      claude-opus-4            → max
#      claude-sonnet-4-5        → high
#      ..."

Status bar indicator

The extension adds a status bar entry showing the current model and level:

Indicator	Level
⚪ model	off
🔵 model	minimal
🟢 model	low
🟡 model	medium
🟠 model	high
🔴 model	xhigh
💜 model	max

The status updates automatically on model change and level change.

Customization

Want different mappings? The extension is designed for easy modification. Edit modelMap in index.ts to add your own patterns, or use the /reasoning reset command to restore defaults after experimenting.

Pattern matching rules

• Patterns are matched as case-insensitive substrings against the model ID
• Entries are checked in order — first match wins
• Put more specific patterns before broader ones
• Optionally filter by provider (e.g. providers: ["anthropic"])

Example custom entries

// Custom provider filter
{ pattern: "my-custom-model", level: "high", providers: ["my-provider"] },

// Broad family match (caught by more specific entries first)
{ pattern: "gpt", level: "medium" },

Development

# From monorepo root
cd packages/pi-reasoning
pi -e .                  # Test locally

Files

File	Role
index.ts	Extension entry — model map, event handlers, /reasoning command
package.json	Package metadata, peer deps
README.md	This file

How it integrates

The extension uses two pi.dev lifecycle hooks:

1. model_select — Fired when the model changes. We look up the new model in our mapping and apply the corresponding thinking level.
2. thinking_level_select — Fired when the thinking level changes (from any source). We keep the status bar in sync.

The pi.setThinkingLevel() API handles clamping to model capabilities, so a max setting on a non-reasoning model is safely ignored.

Related

• pi.dev extensions documentation
• pi.dev models documentation

License

MIT