Pi Guide

A minimal terminal coding agent with configurable providers, sessions and extension packages.

Last verified: 2026-09-22 · How we verify

Coding agentEarendil & contributorsmacOSWindowsLinux

Suitable workflows

Developers building a small, customizable agent workflow around their own tools.

Extensions can expand its permissions and behavior; the default tools can already edit files and run Bash.

Agent workflowCLI surfaceRules / instructionsSkillsBring your own key

This is an evidence-based shortlist, not an exhaustive feature inventory. Omitted capabilities have not been checked for this record.

See official pricing

Free CLI; model-provider or optional service charges. The cost of metered usage or an external agent can be separate from the subscription.

Supported provider API keys or account logins; check provider terms and current compatibility.

Verified families: Claude, GPT.

npm and a supported model provider; the current package is published under @earendil-works.

Installation command · npm

npm install -g --ignore-scripts @earendil-works/pi-coding-agent

Open your project folder in a terminal, then run:

pi

Use /login to choose credentials, or provide the selected provider’s API key through its environment variable.

AGENTS.md supplies context. /model chooses the model; /resume and /tree navigate saved sessions.

Request an explanation of the repository and its checks before allowing a scoped edit.

This example asks for a project summary. Tool permissions still apply; a prompt is not a sandbox.

pi -p "Summarize the project structure without modifying files."

Terminal pair programming with repository maps, editable file context and Git integration.

An open-source coding agent with a terminal UI and configurable model providers.

An open-source agent that connects model providers, MCP tools and repeatable recipes.

Open a small, familiar project and confirm that your essential language tooling works.
Give the agent one bounded change with an observable acceptance criterion.
Review the diff, run the project’s checks and compare the result with your current workflow.
Inspect usage after the task. Estimate a month of similar work, including any separate model or agent bill.