CLI tool for conversing with AI models via OpenRouter. Maintains conversation state across invocations for coherent multi-turn conversations directly from your terminal.

This tool is still in early development and not ready for general use.

Features

• Persistent conversations - State saved between sessions
• Multiple contexts - Separate conversations for different projects/topics
• Plugin system - Extend capabilities with custom tools
• Streaming responses - Real-time output as the AI responds
• Rolling compaction - Automatic context management with intelligent summarization
• Agentic workflows - Built-in tools for todos, goals, and autonomous processing
• Cross-context messaging - Contexts can communicate with each other
• Large output caching - Tool outputs automatically cached with surgical access tools
• Unix philosophy - Only LLM output goes to stdout (pipeable)

Quick Start

Install

cargo install --path .

Configure

Create ~/.chibi/config.toml:

api_key = "your-openrouter-api-key"
model = "anthropic/claude-sonnet-4"
context_window_limit = 200000
warn_threshold_percent = 80.0

Copy example prompts:

mkdir -p ~/.chibi/prompts
cp examples/prompts/*.md ~/.chibi/prompts/

Use

# Simple prompt
chibi What is Rust?

# Pipe content
cat error.log | chibi "explain this error"

# Different contexts
chibi -c coding "Review this function"
chibi -c research "Find info about X"

# See tool activity
chibi -v "Read my Cargo.toml"

Documentation

• Getting Started - Installation and first steps
• Configuration - Full config reference including API parameters
• Contexts - Managing multiple conversations
• Plugins - Creating tools for the LLM
• Hooks - Lifecycle event system
• Agentic Workflows - Autonomous multi-step processing
• CLI Reference - All command flags
• Transcript Format - JSONL format specification

Command Overview

# Contexts
chibi -c <name>       # Switch context
chibi -C <name>       # Transient context (one-off)
chibi -L              # List contexts
chibi -l              # Current context info

# History
chibi -a              # Archive current context
chibi -z              # Compact current context
chibi -g 10           # Show last 10 log entries

# System prompts
chibi -y "prompt"     # Set current context's prompt
chibi -n system_prompt # View current prompt

# Tools
chibi -v              # Verbose (see tool calls)
chibi -p plugin args  # Run plugin directly

See CLI Reference for the complete list.

Example Plugins

See chibi-plugins for ready-to-use plugins:

• read_file - Read file contents
• fetch_url - Fetch web content
• run_command - Execute shell commands (with confirmation)
• web_search - Search via DuckDuckGo
• recurse - Continue processing autonomously
• sub-agent - Spawn sub-agents in other contexts
• github-mcp - GitHub integration via MCP

Storage

~/.chibi/
├── config.toml           # Global configuration
├── models.toml           # Model metadata (optional)
├── prompts/              # System prompts
│   ├── chibi.md          # Default prompt
│   └── reflection.md     # LLM's persistent memory
├── plugins/              # Plugin scripts
└── contexts/<name>/
    ├── context.jsonl     # Conversation history
    ├── local.toml        # Per-context config
    ├── todos.md          # Current todos
    ├── goals.md          # Current goals
    ├── system_prompt.md  # Custom prompt (optional)
    └── tool_cache/       # Cached large tool outputs

License

ISC

Make meow, not rawr