pi-go CLI
A terminal coding agent in Go on Google ADK: several model providers, subagents, LSP, memory, MCP and skills.
- 224 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- The shell tool runs commands with your account permissions.
- The install downloads a binary through a curl script.
Install
Manual install
curl -fsSL https://raw.githubusercontent.com/dimetron/pi-go/main/scripts/install.sh | bashFor macOS and Linux. On Windows use `scripts/install.ps1`. Run `pi verify` after installing.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The shell tool runs commands with your account permissions.
- The install downloads a binary through a curl script.
This is third-party code. Review the repository files before installing.
What it does
`pi` is a terminal coding agent on Google ADK Go. It connects to Claude, OpenAI models, Gemini, Mistral, Grok, Azure OpenAI, OpenRouter, OpenCode and Ollama, and the read, write, edit, shell, grep, find, tree and git tools are restricted to the project directory through `os.Root`. The interface is built on Bubble Tea with Markdown and slash commands, and sessions are stored as JSONL logs with branching, compaction and resume. Model roles (default, smol, slow, plan, commit) are chosen with flags. The process-based subagent system knows the types explore, plan, designer, reviewer, task and quick_task. The LSP client works with Go, TypeScript and JavaScript, Python, Rust and Java, and the Git tools write repository overviews and commit messages through `/commit`. A Unix-socket RPC server for editors, hooks, `.SKILL.md` skills and MCP servers extend the agent, and Memory Palace keeps four memory layers in SQLite with embeddings. Install with the `install.sh` script on macOS and Linux and `install.ps1` on Windows, verified by `pi verify`. Plugins from Claude Code-compatible marketplaces connect as skill bundles.
Who it is for. Developers who need a terminal coding agent with a choice of model provider, subagents and extension through skills and MCP.
Good fit when
- The agent has to switch between Claude, GPT, Gemini and local Ollama.
- A task must be split between explore, plan and reviewer subagents.
- Commits in the project are written with the `/commit` command by change topic.
- An editor connects to the agent through RPC over a Unix socket.
Not a fit when
- You need IDE integration with a graphical interface: the core of the project is the terminal.
- You work on Windows without `bash.exe`: commands have to be written in PowerShell syntax.
- You need a provenance check without running `pi verify` after install.
Example request
Run pi in plan mode and review how authorization is built in this repository, then propose a refactoring.Limitations
The installer checks the archive against the release `checksums.txt`, which does not protect against a substituted release, so run `pi verify` after installing. Tools are restricted to the project directory, but the shell runs commands with your permissions. Windows is supported on amd64, and without `bash.exe` commands are written for PowerShell. The keys of the chosen providers are required and local models go through Ollama. The license is MIT, the last commit dates from November 2026.
How to disable. Delete the `pi` binary from `/usr/local/bin` or `~/.local/bin` (on Windows `pi.exe` from `%LOCALAPPDATA%\Programs`) and the project `.pi-go` directory.
FAQ
How do I install pi-go?
On macOS and Linux `curl -fsSL https://raw.githubusercontent.com/dimetron/pi-go/main/scripts/install.sh | bash`, on Windows `install.ps1` from PowerShell. Then `pi verify`.
Which providers are supported?
Claude, OpenAI, Gemini, Mistral, Grok, Azure OpenAI, OpenRouter, OpenCode and Ollama, local or cloud.
Which subagents exist?
Process-based ones: explore, plan, designer, reviewer, task and quick_task.
How do I connect MCP and skills?
Through `.pi-go/mcp.json` for MCP servers and `SKILL.md` files in `.pi-go/skills`. Plugins from Claude Code-compatible marketplaces install as skill bundles.
What is Memory Palace?
A four-layer memory in SQLite with all-MiniLM-L6-v2 embeddings, a temporal knowledge graph and project and conversation miners.
Discussion
No questions or reviews yet.
Related
CLI toolsA terminal coding agent in Go: GLM, DeepSeek and local Ollama models, 59 built-in tools and MCP
A Pi agent package: code search, LSP, web search, desktop control and confirmations for risky commands.
A Python terminal coding agent in the manner of Claude Code: reads the repository, edits files, runs commands, with memory, sub-agents and MCP
A Rust coding agent with a native GUI, a terminal mode, an ACP mode for Zed and an MCP server
Over twenty skills for coding agents: MCP and hooks, plan mode, a bug-fix protocol, prompting and a bash safety hook
Native interface for the pi agent in Rust and GPUI: project sessions, a conversation tree, a changes panel and plugins.
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| pi-go (this card) | no | 224 | |
| Crush | no | 29,067 | Crush remains a general terminal agent, while pi-go adds process-based subagents, LSP, editor RPC and Memory Palace. |
| OpenCode | no | 226,389 | OpenCode is written in TypeScript, while pi-go is built on Go and Google ADK and installs as one binary. |
README in short
The pi coding agent in Go on Google ADK for the terminal: nine model providers, subagents, LSP, Memory Palace memory, MCP and skills.