pi-effort

Extension for pi: small command surface for reasoning effort and OpenAI/Codex fast mode.

Claude Fable reviewed 0.0.7 on 2026-08-15. See SECURITY.md.

Commands

Exactly two slash commands are exposed:

/effort [min|minimal|low|medium|high|xhigh|max|off|default]
/fast [on|off]

Bare /effort opens a picker with the current model's levels; Enter applies, Esc cancels.

No /effort show, /effort options, /effort fast, or /fast status. Bare /fast toggles the current fast-mode setting.

Effort

/effort accepts the current model's supported reasoning levels plus two adaptive aliases:

off and default differ:

• /effort off turns thinking off. Pi sends an explicit "no thinking" field, for example reasoning: { effort: "none" } to OpenAI or thinking: { type: "disabled" } to Anthropic.
• /effort default asks the provider for the model's own default effort. Pi has no such level, so the extension runs Pi at off and removes those "no thinking" fields from each request. The footer shows think:default. Any other level ends it.

Models that cannot turn thinking off (their thinkingLevelMap sets off to null, such as gpt-5) offer neither.

Examples:

/effort min
/effort medium
/effort max
/effort off
/effort default

Fast mode

Fast mode is the latency/service-tier knob. When enabled, pi-effort adds service_tier: "priority" only for OpenAI, Codex, and Azure OpenAI requests whose model id starts with gpt-5. That tier is billed at a higher token price. The setting is global and persists until /fast off. A session that starts with it on prints a warning.

/effort calls Pi's setThinkingLevel. Pi saves that as defaultThinkingLevel for later sessions.

/fast      # toggle
/fast on   # force on
/fast off  # force off

Fast mode persists in ~/.pi/agent/settings.json under:

{
  "pi-effort": {
    "fastMode": true
  }
}

Footer status

For compact powerline footers, the extension publishes these status keys:

• pi-effort-thinking — think:<level>
• pi-effort-fast — fast only when fast mode is enabled and applies to the current model

Display settings

The extension also sets the working message to Working (<level> effort).... Turn off either display in ~/.pi/agent/settings.json:

{
  "pi-effort": {
    "showModelEffortInWorkingMessage": false,
    "showModelEffortInFooter": false
  }
}

• showModelEffortInWorkingMessage: false — the extension stops setting the working message and clears the one it set before, so Pi's default message returns.
• showModelEffortInFooter: false — the extension clears both footer status keys.

Both default to true. Only an explicit false hides a display.

Keyboard shortcut

Ctrl+Shift+E cycles through the current model's reasoning levels.

Per-model effort

The extension remembers the last level you chose for each model, and restores it when you switch back to that model with /model or model cycling. Example: xhigh on a GPT model, medium on Opus. Switching from GPT to Opus and back gives medium, then xhigh again, instead of the level Pi carried over and clamped.

• A level counts as chosen when you set it with /effort (including default), Ctrl+Shift+E, or Pi's own thinking selector. The level Pi picks itself when a model cannot run the previous one is not remembered.
• A pinned level in enabledModels or --models (for example anthropic/*:high) wins when cycling.
• A remembered level the model no longer supports is skipped.
• Startup and resumed sessions keep the level Pi chose (--effort, --thinking, a :level model suffix, or the saved default). Restore happens on model switches only.

The levels live in ~/.pi/agent/settings.json, keyed by provider/model-id. Delete a key to make the extension forget that model:

{
  "pi-effort": {
    "modelEffort": {
      "openai/gpt-5.4": "xhigh",
      "anthropic/claude-opus-4-6": "medium"
    }
  }
}

CLI flag

pi --effort max
pi --effort min
pi --effort high
pi --effort off
pi --effort default

The flag uses the same values as /effort.

Install

Requires Pi 0.84 or newer (@earendil-works/pi-coding-agent).

pi install npm:@machine-made/pi-effort

Until npm has the latest tag, install from git:

pi install git:github.com/machine-made/pi-effort

Verify what Pi is loading:

pi list
npm list -g --depth=0 pi-effort

The durable Pi setting is the package entry in ~/.pi/agent/settings.json:

{
  "packages": [
    "npm:@machine-made/pi-effort"
  ]
}

Local development

For a one-off local test from this checkout:

npm install
pi -e ./index.ts

For a longer local-path development install:

pi remove npm:@machine-made/pi-effort
pi install "$(pwd)"

After publishing a new version, switch back to the npm package:

npm publish
pi remove "$(pwd)"
pi install npm:@machine-made/pi-effort

Verification

npm run check
npm test
npm pack --dry-run

Repo structure

index.ts        Pi extension entrypoint
effort.ts       Parsing, settings, and model capability logic
package.json    Package metadata and Pi manifest
tsconfig.json   TypeScript configuration

License

Apache-2.0

Disclaimer

This software is provided "as is", without warranty of any kind. To the extent permitted by law, the authors and contributors are not liable for any damage, loss or claim arising from its use or misuse. You are responsible for how you use it and for following the laws and rules that apply to you. The full terms are in LICENSE.