ZEUS - Local AI coding agent for your desktop

by•
ZEUS is a desktop coding agent for Linux powered by your local Ollama models. Give it a task or attach a spec (.md, PDF, Word…) and it works in rounds until the result is verified. You approve every change or set to auto approve mode. Nothing leaves your machine. Open source.

Add a comment

Replies

Best
Maker
📌

Hi Product Hunt 👋 I built ZEUS because I wanted a coding agent that works like the cloud ones but runs entirely on my own machine, with my own models, and doesn't claim it's done when it isn't.

I built ZEUS solo, using Claude Code.

What it does?

  • Point it at a project folder and pick any Ollama model. It reads, edits and runs code with tools, and every write or command goes through a permission mode you choose. Approvals show up inline with the diff or the exact command.

  • Missions: attach a spec (a .md, or a PDF, Word, PowerPoint or Excel file, converted locally with Microsoft's MarkItDown). ZEUS works in rounds, then a reviewer checks each acceptance criterion against evidence. Objective checks overrule the model: tests must exit 0 and are re-run to confirm, and named files must exist. Anything unmet goes back to the agent.

  • Model-aware: it prepares each request for the specific model (sampling, thinking mode, tool calling, context size). Auto mode picks a coding or a reasoning model per prompt.

  • A Models tab to browse and download Ollama models and see GPU/CPU use live.

Honest limits (v0.1.0)

  • Linux only for now (a .deb, tested on Ubuntu 24.04).

  • Quality depends on your model. A 9B model finished a multi-file mission for me; a 7B one often stalls (ZEUS tells you rather than faking success).

  • Commands run as your user. ZEUS gates them but doesn't sandbox them.

It's MIT-licensed and early. I'd love your feedback on which models and tasks work (or don't) on your hardware. What would you want from a local agent?