fly through
code.

CHYRD is an AI coding agent in your terminal: one interface for 34 providers and 169+ models - hundreds more via live discovery - with MCP servers, per-turn undo, subagents, skills, and an unattended loop mode. Free to install and use.

$ pipx install chyrd

read the docson PyPI

        ,___,
       ( O,O ) 
       (  v  )
---------"-"---------

what it looks like

A real CHYRD session: user request, streamed reply, read and edit tool cards with a colored diff, the mascot owl, and the status bar with tokens, cost, and mode
A real session, exported by CHYRD itself from a headless run - not a mockup. What you install is what you see.

what's inside

  1. 01

    34 providers, 169+ models

    Anthropic, OpenAI, Google, OpenRouter, Groq, GLM, Kimi, and twenty-seven more - and with a key set, providers like NVIDIA and OpenRouter list their FULL live model catalogs (hundreds) at startup.

  2. 02

    MCP servers

    declare any MCP server in config; every tool it advertises becomes a CHYRD tool, gated by the same permission system.

  3. 03

    undo, per turn

    every file mutation is journaled before it happens. /undo reverses the whole last turn - even a deleted directory.

  4. 04

    permission modes

    normal, accept-edits, plan, bypass - cycled live with shift+tab. The approval dialog shows the exact diff or command first.

  5. 05

    subagents and skills

    nested focused agents plus 15 expert playbooks, both defined as plain markdown files you can write yourself.

  6. 06

    loop mode

    chyrd loop runs unattended - one focused change per pass, and it halts itself after three failed passes.

  7. 07

    sessions, headless, export

    conversations persist as JSONL and resume exactly; chyrd run scripts it for CI; /export writes a transcript.

  8. 08

    your shell, your context

    prefix a message with ! to run a command yourself; the output is folded into the model's context automatically.

you decide how much to trust

four permission modes, cycled live with shift+tab.

  • normalask before every mutating action - write, edit, shell, fetch.
  • accept-editsfile writes and edits auto-approved; shell and fetch still ask.
  • planread-only. mutating tools are refused with guidance; present a plan.
  • bypasseverything auto-approved. move fast; nothing prompts.

bring whatever key you already have

Anthropic / OpenAI / Google / OpenRouter / Vercel AI Gateway / Poe / Groq / DeepSeek / xAI / Mistral / Together / Fireworks / Moonshot / Cohere / Qwen / Z.AI / MiniMax / Perplexity / AI21 / Inception / DeepInfra / Novita / Nebius / SiliconFlow / Hyperbolic / GitHub Models / Cerebras / NVIDIA / SambaNova / Hugging Face / Chutes / Ollama / LM Studio / any OpenAI-compatible URL

green = a genuinely usable free tier. Ollama and LM Studio run keyless on localhost. Keys come from environment variables or /keys inside the app.

compared, honestly

the others are excellent tools; this is exactly where they differ. ● yes, ○ no.

CHYRDOpenCodeGemini CLIClaude Code
Providersallmanyoneone
Local models (Ollama)●●○○
MCP servers●●●●
Skills library●○○●
Subagents●●○●
Permission modes●●●●
Undo file changes●○○●
Loop mode●○○○
Pricefreefreefreepaid

providers: all = every major API plus local and custom endpoints; many = a broad hosted set; one = a single vendor (Google / Anthropic).

install

pipx install chyrd        # recommended: isolated, on your PATH
pip  install chyrd        # into the current environment
uv   tool install chyrd   # if you use uv

Python 3.10+. Windows, macOS, Linux. No key at all? Serve a model with ollama on localhost:11434 and just run chyrd - fully keyless. To use the Claude Code engine as a backend (chat-only - CHYRD's tools stay off there):

pipx install "chyrd[claude]"
chyrd --model claude-agent/sonnet     # uses ANTHROPIC_API_KEY

full documentation - keys, keybindings, configuration, every command.