StarNet
StarNet is a local-first desktop application that runs real AI agents inside a pixel-art space station, where rooms represent capability-scoped teams and the layout you build is the actual multi-agent workflow that executes.
StarNet is a desktop harness for creating AI agents, organizing them visually, and running them with real language models, tools and costs rather than a simulation. Each room in the pixel-art station is a capability-scoped team, each hallway is an authorized handoff between agents, and each placed object is a real capability grant, so the layout a user draws is the workflow the agents execute.
Users can run several agents concurrently, each with its own workspace, transcript, memory and bounded permissions, and bring their own model access through an OpenRouter key or by signing in with Anthropic, OpenAI or Google accounts, with keys stored in the OS keychain. Agents can be reached from Telegram, Discord, Slack, Signal or Matrix, can keep working unattended within an adjustable, logged permission leash, and can be extended with Model Context Protocol connectors.
Finished work is delivered as real files in an outbox rather than left in chat history, and spend, budgets and run history are recorded to disk.