Skip to content

Guide

Cookbook

Reference

Concepts

01 — Installation

Get a working reyn install in under 5 minutes.

Prerequisites

  • Python 3.11+
  • A LiteLLM-compatible model endpoint (OpenAI, Gemini via Google AI, Anthropic, or a local proxy like LiteLLM Proxy)

Install

git clone https://github.com/tya5/reyn.git
cd reyn
python -m venv venv
source venv/bin/activate
pip install -e '.[dev]'

The reyn CLI is now on your PATH.

Configure a model

reyn picks the model from reyn.yaml. The shipped default uses Gemini via a LiteLLM proxy. To use a different provider, edit the llm.models map:

# reyn.yaml
llm:
  model: standard
  models:
    light:    openai/gpt-4o-mini
    standard: openai/gpt-4o
    strong:   anthropic/claude-3-5-sonnet-20241022

The shorthand <value> above is a literal model string (= contains /). Reyn ships no built-in model catalog — light/standard/strong are reyn's own tier names, but what each one points to is entirely this mapping's job; a dict form with extends is also available for cost variants. See reference/config/reyn-yaml.md for details.

Then export your provider's API key, e.g.:

export OPENAI_API_KEY=sk-...

litellm resolves your provider's own variable automatically — see litellm's provider docs for the name yours expects.

Never commit API keys

Keys belong only in environment variables. reyn.yaml is checked in; put proxy URLs in reyn.local.yaml or ~/.reyn/config.yaml (gitignored).

Initialize a project

In your working directory:

reyn init

This creates reyn.yaml and reyn.local.yaml.example if they don't exist.

Project context (optional)

Drop an AGENTS.md in your project root and Reyn injects it into every session as project-specific context (conventions, architecture notes, do's and don'ts). AGENTS.md is the cross-tool standard that Claude Code, Codex, and opencode also read, so a project you already share with those tools works as-is — no Reyn-specific file needed (a legacy REYN.md is still honored as a fallback). To pin a different file or turn it off, see project_context_path.

Verify

reyn skills          # lists stdlib + project + local skills
reyn run direct_llm "reyn is a workflow OS for LLMs."

If the second command produces a summary and exits cleanly, you're ready for 02 — Chat mode.

Troubleshooting

  • reyn: command not found — your venv isn't active. source venv/bin/activate.
  • AuthenticationError — the API key env var isn't set, or doesn't match the model in reyn.yaml.
  • Proxy connection refused — start your LiteLLM proxy, or remove api_base from reyn.local.yaml.

On this page