tkellogg/lanius

Lanius is a local control plane for AI work, built in Rust. The idea behind it is that a human should stay in charge of attention, budget, and intent, while different AI models, scripts, coding tools, and background workers each handle the part of the job they are suited for. Instead of building one giant assistant that tries to do everything, Lanius makes it easy to send cheap tasks to a cheap model, quick interactions to a fast model, and important decisions to a stronger model, all coordinated through one system running on your own machine. You install it with Cargo, Rust's package manager, then run a couple of commands to set up a local root folder and start the service. Once running, it opens a web interface where you can chat with AI profiles, configure model providers such as Anthropic, inspect coding sessions, manage installed packages, and answer questions the system is waiting on you to decide. Everything you can do from the web interface can also be done from the command line, since the interface itself uses the same underlying commands. A central feature is running coding tools like Claude Code, Codex, or opencode through Lanius so their work becomes visible, addressable, and something you can come back to later. You can launch these tools directly on a project, run a task headlessly without opening an interactive window, watch a session live, or pull up a full report of what a session did, including which files changed. Workers can also start other workers or send each other messages, so one AI can plan a task while another implements it and a third checks the result, with a human only stepping in for decisions that actually need a human. Lanius uses a mailbox model for communication: any part of the system, including scripts and webhooks, can send messages if it has permission to do so, and you can inspect your inbox and answer pending questions. The system can be extended through packages, which may bundle instructions, memory, background processes, scheduled jobs, and new tools. The project is licensed under Apache 2.0, a permissive license that allows free use, modification, and distribution, including commercial use.