Chibi is a composable building block for LLM interactions — not an agent framework. It provides persistent context storage, a plugin and hook system that exposes the full request lifecycle, communication primitives for multi-agent coordination, and a programmable agentic loop.
Plugins have full access to chibi's lifecycle hooks — they can modify or replace core behaviours (like context assembly, prompt construction, tool filtering, permissions), spawn sub-agents with custom home directories, orchestrate fleets of chibis across different models and system prompts, and implement entirely new coordination patterns. The plugin layer has no ceiling.
Early development — not yet stable.
export CHIBI_API_KEY=your-openrouter-key
git clone https://github.com/emesal/chibi.git
cd chibi && just install
chibi nyaaa~Requires just and a Rust toolchain. Cargo fetches all dependencies automatically.
Get a free API key at openrouter.ai (no credit card needed), then:
chibi What is Rust? # Simple prompt
cat error.log | chibi "explain this" # Pipe content
chibi -c project "Review this function" # Named context
chibi -v "Read my Cargo.toml" # Verbose (show tool use)Contexts persist across invocations. Switch with -c <name>, list with -L.
Persist your key and pick a model in ~/.chibi/config.toml:
api_key = "your-key-here"
model = "anthropic/claude-sonnet-4" # default: free-tier agentic presetAll fields are optional. See Configuration for the full reference.
- Getting Started — Installation and first steps
- Configuration — Full config reference
- Contexts — Managing conversations
- Plugins — Creating tools for the LLM
- Hooks — Lifecycle event system
- MCP Servers — Using MCP-compatible tool providers
- Virtual File System — Sandboxed shared file space for contexts
- Agentic Workflows — Autonomous processing
- CLI Reference — All flags and commands
- Images — Terminal image rendering
- Markdown Themes — Customising colour schemes
- Transcript Format — JSONL format spec
Example plugins: chibi-plugins
ISC
Make meow, not rawr