agentdesktop inventories every agent, model, MCP server and skill on a machine, applies central policy, and issues agents short-lived per-device credentials.
Yes, agentdesktop is open source under the Apache-2.0 license.
agentdesktop is free to use.
Yes, agentdesktop can be self-hosted (the source is available under the Apache-2.0 license).
Open-source visibility and control layer for AI tools across a desktop fleet (Apache-2.0, Rust), backed by Solo.io: it discovers every harness, model, MCP server and skill on a machine attributed to a device and user, applies centrally-declared policy that reconciles drift, and treats agents as non-human identities issued short-lived, just-in-time credentials bound to the device, user and tool. Runs standalone from local YAML on one workstation, or in fleet mode via an MDM-installed daemon wired to your existing identity provider and LLM gateway.
This is the fleet-scale version of the agent-security problem this registry keeps flagging. Geiger shows you what one machine granted; agentdesktop does that across every laptop your team runs agents on, adds policy that reconciles drift, and swaps long-lived keys for short-lived credentials scoped to a device, user and tool. For a GTM or RevOps team where three people now run agents against the CRM and the sending domain, that is the difference between hoping everyone configured it right and being able to see and set it centrally. The honest read: it is a young repo with a serious company behind it, the standalone mode is what you can try this week, and the identity story is genuine infrastructure work, not a checkbox.