Guides

agentty guides

Practical, copy-pasteable walkthroughs for getting the most out of agentty.

• How to cut your coding agent's token usage with retrieval (RAG)Coding agents waste tokens by sending your whole repo to the model. Learn how agentty's built-in retrieval sends only relevant code — cutting context by 80%+.
• How to run a sandboxed AI coding agent (safe by default)AI agents run shell and build commands — that's risky. Learn how agentty sandboxes every command by default with bubblewrap, and how to air-gap a session over SSH.
• How to run an AI coding agent completely locally with OllamaStep-by-step: run a fully local AI coding agent with agentty + Ollama — no cloud, no API keys, no data leaving your machine. Works offline.
• How to use Claude from the terminal (with your Pro/Max plan)Use Claude as a coding agent from your terminal with agentty — sign in with your existing Claude Pro/Max subscription, no extra API cost. Install in one line.
• How to use Groq for blazing-fast coding with agenttyUse Groq's ultra-fast inference as your coding agent backend with agentty — step-by-step setup for low-latency, high-throughput terminal coding.
• How to use OpenRouter as your coding agent backendRoute dozens of models through one API: use OpenRouter with agentty for flexible, cost-optimized terminal coding. Step-by-step setup.
• The best terminal AI coding agent in 2026A rundown of terminal-first AI coding agents in 2026 — and why agentty (native C++26, one static binary, multi-provider, sandboxed) is a top pick for the command line.
• What is an MCP terminal client? (and how to use one)MCP (Model Context Protocol) lets AI agents use external tools. Learn what an MCP terminal client is and how agentty connects to MCP servers from the command line.
• What is code RAG? Retrieval-augmented generation for coding agentsCode RAG means retrieving only the relevant parts of your repository for an AI coding agent instead of sending everything. Here's how it works and why it matters.