Built on Google ADK Go

Your AI coding agent, in the terminal

Multi-provider LLMs, sandboxed tool execution, persistent memory, and a beautiful TUI. All from a single binary.

View on GitHub Quick Install

$ curl -fsSL https://pi-go.sh/install | bash ⎘

pi-go terminal UI — minimax-m2.7:cloud model with Ollama

Everything you need in one binary

A fully-featured coding agent that respects your terminal workflow

Multi-Provider LLM

Claude, GPT, Gemini, and Ollama for local models. Switch providers mid-conversation with a slash command.

Sandboxed Tools

File read/write/edit, shell execution, grep, find, tree, and git — all restricted to the project directory via os.Root.

Memory Palace

4-layer contextual memory with SQLite, semantic embeddings, temporal knowledge graph, and project miners.

LSP Integration

JSON-RPC client for Go, TypeScript, Python, and Rust with auto-format and diagnostics hooks.

Subagent System

Process-based multi-agent orchestration with types: explore, plan, designer, reviewer, and task runners.

Session Branching

JSONL append-only event logs with branching, compaction, and resume. Fork conversations and switch paths.

A Anthropic

O OpenAI

G Gemini

L Ollama

Up and running in seconds

One command to install. One binary to run.

terminal

# Install pi-go
$ curl -fsSL https://pi-go.sh/install | bash

# Or with go install
$ go install github.com/dimetron/pi-go/cmd/pi@latest

# Start interactive mode
$ pi

# Select a model
$ pi --model claude:sonnet
$ pi --model openai:gpt-4o
$ pi --model gemini:gemini-2.5-pro
$ pi --model ollama/qwen3.5:latest
$ pi --model minimax-m2.7:cloud

# Use model roles
$ pi --smol          # fast, cheap model
$ pi --slow          # most capable model
$ pi --plan          # planning-oriented model
        

slash commands

/model         Switch model mid-conversation
/session       List and switch sessions
/branch        Create a conversation branch
/commit        AI-generated conventional commits
/plan          Start PDD planning session
/run           Execute a spec with task agent
/memory        Memory Palace commands
/audit         Scan skills for Unicode threats
/agents        Show running subagents
        

Clean architecture

Modular Go packages with clear separation of concerns

User input  →  CLI  →  Agent  →  LLM provider  →  Tool calls  →  Response  →  TUI
                         ↓          ↓                 ↓
                  Session store    Palace            LSP servers
                  (JSONL events)   (memory, KG,      (format, diagnostics)
                                    search)

⚙

Agent & CLI

Cobra CLI, ADK runner, retry logic, output modes (interactive, print, JSON, RPC).

🛡

Sandboxed Tools

Read, write, edit, bash, grep, find, git, LSP — all scoped to project root.

🧠

Memory Palace

SQLite drawers, FTS5 search, semantic embeddings, temporal knowledge graph.

🔀

Session & Branching

JSONL event logs, session persistence, branching, compaction, and resume.

🤖

Subagents

Process-based orchestration with concurrency pool and typed agents.

🔌

Extensions

Shell hooks, skill files, MCP servers, and security auditing.