Raven

Why we included this project

Raven starts from a different assumption than most agent tools: the harness around the LLM is the real product, not the model. You run it from the terminal, and it keeps user memory, agent memory, and world knowledge separate and durable across sessions instead of cramming everything into one context window. For teams that want an agent to become part of their daily workflow, the self-improving loop is the draw: repeated tasks harden into reusable skills and templates rather than vanishing into chat history. The opt-in deep research mode pulls in multi-source investigation when a task genuinely needs it, and the messaging gateways (Telegram, Slack, Discord, WhatsApp, and others) let you reach the same harness from whichever chat app you already live in. It is a practical pick for developers who want an installable, long-lived agent they can extend and steer.