pie

A fast, minimal AI coding agent in Rust. Any OpenAI-compatible provider, persistent sessions, skill-based subagents, and sandboxed shell execution.

Quick start

# Interactive mode — just start talking
pie

# Continue your last conversation in this directory
pie -c

# Pipe a question
echo "what does src/main.rs do?" | pie --md

# Use a specific model
pie -m claude-sonnet-4-20250514

Features

• Persistent sessions — conversations saved per directory, resume with pie -c
• Any provider — works with OpenAI, Anthropic, Groq, Ollama, or any OpenAI-compatible API
• Skills & subagents — markdown-based skills from agentskills.io, auto-loaded from queries
• Sandboxed commands — all shell commands run through srt (OS-level isolation, no containers)
• Streaming TUI — real-time tool calls, markdown rendering, command history
• Scriptable — --json and --md flags for single-shot mode

Usage

# Interactive (default)
pie

# Continue last session
pie -c

# Single-shot output
pie --md "explain this function"
pie --json "list files"   # pipe into jq, etc.

# Use a specific skill
pie "/explore summarize this repo"

Interactive commands

Input	Action
<query>	Ask a question (auto-detects skills)
/<skill> <query>	Use a specific skill
?	Show help
Ctrl+C	Abort stream / quit

Configuration

# .pie/sandbox.json — sandbox restrictions
# .pie/skills/<name>/SKILL.md — custom skills
# .pie/agents/<name>.md — custom subagents
# AGENTS.md — project-level instructions

Flag	Description
-m	Model name
--base-url	API base URL
--api-key	API key
-c	Continue last session
--md	Markdown output (single-shot)
--json	JSON output (single-shot)
-d	Debug logging

Install

Homebrew (macOS / Linux)

brew tap dineshdb/pie https://github.com/dineshdb/pie
brew install dineshdb/pie/pie

From source

cargo build --release

License

MIT