Pi Coding Agent

Pi Coding Agent is an open-source terminal coding agent from Earendil Inc. It is distributed as the npm package @earendil-works/pi-coding-agent and runs with the pi command inside a project directory. Pi provides an interactive terminal UI, project-aware file and shell tools, automatic session saving, model switching, and extension points for teams that want to shape the agent around their own workflows.

Top features

• Terminal-first workflow: start pi in a repository and prompt it to inspect files, edit code, write files, or run shell commands in the current working directory.
• Configurable tools: by default, Pi gives the model read, write, edit, and bash; read-only tools such as grep, find, and ls can be enabled or allowlisted through tool options.
• Multi-model support: authenticate through subscription logins or API keys, switch models with /model or Ctrl+L, cycle scoped models with Ctrl+P, and set thinking levels from the CLI or UI.
• Session tree and branching: sessions are stored as JSONL files under ~/.pi/agent/sessions/ and support resume, fork, clone, tree navigation, compaction, export, and sharing flows.
• Project context and skills: Pi loads AGENTS.md or CLAUDE.md from global, parent, and current directories; it also supports prompt templates and Agent Skills-style capability packages.
• TypeScript extensions and packages: extensions can add commands, tools, UI, hooks, event listeners, and custom compaction behavior, and can be installed locally or globally through Pi packages.
• Programmatic integration: the included SDK exposes agent sessions, tools, resource loading, events, settings, and run modes; --mode rpc provides JSON-RPC over stdin/stdout for language-agnostic subprocess integrations.

Use cases

• Run interactive coding sessions that can edit a repository and execute project commands.
• Review code in read-only mode by limiting tools to read, grep, find, and ls.
• Build repeatable team workflows with AGENTS.md, prompt templates, skills, and local extensions.
• Resume, branch, or export long development sessions for review and debugging.
• Embed Pi in another Node.js application through the SDK, or integrate from another language via JSON-RPC.