fast-agent

Why we included this project

fast-agent is a terminal-first way to work with LLMs: `!` drops you into a shell mid-session, so you can run build commands and manage skills through the built-in registry without switching tools. It connects to MCP servers over stdio or streamable HTTP (OAuth included), and it can expose itself to ACP clients, which lets other tools reuse it as a shared engine. For people building agent workflows, the standout pieces are the subagents, compaction strategies, and the ability to switch models across providers, including a helper that generates the right config for llama.cpp. It is a hands-on toolkit rather than a turnkey product: expect to spend time in the docs and configuration, but the trade-off is that you can iterate on agent behavior quickly without writing a framework from scratch.