Pi Coding Agent
AIA terminal-based, extensible TypeScript coding agent and toolkit for building agentic developer workflows.
-(0 Reviews)
Free Available
Starting from Free
Pi Coding Agent
AI
A terminal-based, extensible TypeScript coding agent and toolkit for building agentic developer workflows.
-(0 Reviews)
Free Available
Starting from Free
About Pi Coding Agent
Pi Coding Agent is an open-source, terminal-first coding agent and toolkit delivered as a TypeScript monorepo. It provides a unified LLM provider layer, an agent loop, a production-ready coding agent runtime with built-in file and shell tools, session persistence (JSONL), and context compaction to keep conversations within model windows. Pi is designed for extensibility via TypeScript extensions, custom tools, skills, prompts and themes, and offers both interactive TUI and Emacs frontends. It runs locally or against hosted providers, includes guidance for containerization and permission boundaries, and is distributed under the MIT license for self-hosting and modification.
Screenshots
+
Key Features
Unified LLM Providers: Abstracts communication with multiple LLM providers and supports provider authentication via /login or environment variables (e.g., ANTHROPIC_API_KEY), letting the agent switch between hosted and local models.
Built-in Coding Tools: Ships a production-ready set of built-in tools for file operations, search (grep/find), shell execution (bash), read/write/edit, and sharing code/snippets to streamline common developer tasks inside the agent loop.
Session Persistence & Compaction: Persists full session history to JSONL files and performs automatic in-memory compaction (summarization) when context approaches model window limits while preserving full logs on disk.
TypeScript Extensibility: Extension system and resource loader for adding custom TypeScript extensions, skills, prompt templates, themes, and pi packages to extend agent capabilities and add custom tools.
Programmatic SDK: Exposes programmatic APIs (createAgentSession, createAgentSessionRuntime, InteractiveMode, SessionManager) to embed agent sessions into other tooling or CI environments.
Terminal UI & Editor Integration: Provides a terminal UI (pi-tui) and an Emacs frontend with keybindings, syntax highlighting, and navigation designed for interactive coding workflows.
Security & Containerization Guidance: Explicitly documents permission model (runs with user/process permissions) and offers recommended containerization/sandbox patterns for stronger isolation in production or CI.
Unified LLM provider layer (pi-ai) abstracting multiple model providers and authentication (supports API keys and /login flows)
Agent orchestration loop (pi-agent-core) enabling tool calling and turn management
Production-ready coding agent runtime (pi-coding-agent) with built-in tools (file ops, shell commands, read/write/edit/grep/find/ls, etc.)
Programmatic SDK: createAgentSession, createAgentSessionRuntime, createAgentSessionServices, InteractiveMode, SessionManager and factory hooks
Terminal UI (pi-tui) and CLI entrypoint (pi) for interactive sessions
Session persistence to JSONL + automatic context compaction to manage long histories and model context windows
Extensibility via TypeScript extensions, custom tools, skills, prompts, themes, and resource loaders (DefaultResourceLoader)
Integrations: Emacs frontend, GitHub Action for running Pi in CI/CD, examples and extension patterns in repo
Security / deployment guidance: no built-in permission sandboxing (runs with user permissions) — recommends containerization/sandboxing patterns
Supply-chain hardening practices for npm (save-exact, lockfile policies, audit commands) and release tooling
Use Cases
Interactive Terminal Pair-Programming: Use Pi in a project directory to iteratively write, refactor, and test code using built-in file and shell tools without leaving the terminal.
Custom Agent Frameworks: Build bespoke agent behaviors and pipelines by composing pi-agent-core, adding custom tools or skills, and orchestrating multi-agent workflows (e.g., planner/builder/reviewer chains).
CI/CD Automation: Run Pi inside CI (via community GitHub Actions) to automate code generation, PR descriptions, or repository maintenance tasks as part of build pipelines.
Local Model & Hybrid Deployment: Connect Pi to local LLMs (Ollama, vLLM, etc.) or hosted providers, enabling offline or hybrid workflows for sensitive code or data while following containerization recommendations.
Embedding in Editors and Integrations: Integrate Pi with Emacs or other editor workflows to provide chat-driven code navigation, edits, and file exploration from within the editor.
Open Source Session Sharing & Research: Share anonymized OSS agent sessions to improve agent tooling and model behavior by contributing real-world agent interactions and failure/fix examples.
Interactive terminal coding assistance and REPL-style development sessions
Automating repository tasks and review workflows in CI/CD using the pi GitHub Action
Embedding a coding assistant into developer tools or interfaces (e.g., Emacs integration)
Building custom agentic workflows and tools that combine shell, file, and API operations
Running local or remote LLM-backed coding agents (supports local LLM configuration and provider fallbacks)
Prototyping or shipping production agent systems with session persistence and context compaction
Explore more AI Ai Agents tools
Browse all Ai Agents tools →Browse by use case: Code Generation
Compare Pi Coding Agent: vs Daemons by Charlie Labs · vs Relay · vs Backplanes Spotlight · vs AGNT.Hub