starnet: Visualize AI Agent Teams as a Pixel-Art Space Station

A local-first desktop harness that places AI agents in a pixel-art space station to perform real tasks. Rooms function as permission-scoped teams, corridors as approved data transfer paths, and placed objects as actual functional permissions. The layout drawn by the user becomes the workflow executed by the agents.

This is not a simple simulation; it involves actual model calls and tool execution. You can directly connect API keys from OpenRouter or major LLM providers, or run local models via Ollama. Integrate with external channels like Telegram, Discord, and Slack to issue remote instructions, with work outputs saved as actual files in the OUTBOX rather than chat logs.

Each agent has an independent workspace, memory, and budget limit, enabling concurrent execution. It supports Task Briefs, which ask clear questions instead of guessing when instructions are ambiguous, and Night Shift mode, which continues work within limited scope overnight. All consumption records and execution histories are permanently stored on disk for transparent management.