Long is a self-hosted LLM agent runtime built on Elixir/OTP. Point it at any provider (OpenAI, Anthropic, Google, DeepSeek, local Ollama — 20+ via ReqLLM) and talk to it through a built-in web UI: it runs a ReAct loop with tools, keeps four tiers of memory, schedules recurring tasks, searches and reads the real web, and answers your WeChat and Telegram DMs — all from a single process.

What sets it apart is that the agent’s entire capability layer is one introspectable GraphQL tool over its own data, instead of a hand-maintained pile of bespoke tools — the model already speaks GraphQL, so it discovers and operates its own memory, schedules, and sessions on its own.

And it installs like a CLI tool, not server infrastructure: one self-contained binary (the Erlang VM is bundled), everything under ~/.long/, no external services to provision (just SQLite + the filesystem), userspace autostart, LAN-first defaults — so getting it onto a spare Mac mini or Linux box is curl | bash plus an API key, and it runs like an appliance from there.