CoeOS Desktop
More possibilities: an agent with the whole machine. Your files, your shell, your environment — and every model call still through your CoeOS box.
CoeOS Desktop is the agentic harness of CoeOS, for macOS, Windows and Linux. Where the web app works within guarded multi-agent limits, the desktop gives the agent full access to your machine. It updates itself.
The model always goes through the box
Section titled “The model always goes through the box”The desktop only talks to your instance’s box: models, web search and page reading go through it. Same routing by skill, same confidentiality rules as the web app. A build check verifies that no other part of the code opens a network connection.
The agent
Section titled “The agent”- A root agent plans into a shared task list and delegates to sub-agents running in parallel, in the foreground or the background, under run budgets (steps, tokens, sub-agents, depth, concurrency).
- Machine tools: files anywhere, and the user’s own shell with its environment. No sandbox.
- Agents in Markdown: replace the shipped ones or add your own. A
decisionagent sorts and classifies lists with a decision model. - One model per agent, if you want: a box model for open work, an axis of the Local table for confidential work, or “CoeOS decides” by default.
Permissions
Section titled “Permissions”- Ask (default): every command and every write waits for your approval.
- Grant all: the agent acts without asking.
The mode switches live. Every turn keeps track of its writes: you see the diff and can undo them.
Code mode
Section titled “Code mode”A plan, reviewed in parallel by three agents (direct, alternative, sceptic); a disagreement comes back to you; only your GO approves the plan; then execution by dependencies and a code review. The model cannot approve its own plan.
Confidentiality
Section titled “Confidentiality”- Every request carries its class. A run starts open by default; Confidential is chosen in the composer or in Settings, and the box then serves the models of its Local table.
- A conversation that sent a confidential turn never goes back to open.
- Memory and skills are split by class: an open run never reads what came from a confidential run.
- Every MCP server carries a class; the knowledge base (OdyRAG) is kept out of open runs by default.
Memory and skills
Section titled “Memory and skills”- Memory: at the end of a run, the agent writes down what is durable. Import an existing memory (file, folder, zip); it is searched locally.
- Skills: your Claude Code skills work as they are. A skill written by the agent only loads after you approve it.
Voice (Jarvis)
Section titled “Voice (Jarvis)”Spoken conversation is a deliberate exception: audio goes to Google (Gemini Live), with a short-lived token issued by the box. Turning voice on switches the conversation to open and disables confidential until it stops.