Skip to content

Device farm

A shared device farm (an organizer daemon placing agent requests onto host daemons that dial out to it, observed from the Desktop App) is a proposed design. None of it is implemented: every daemon runs in the single deployment mode it has today, on a laptop, in CI, or on a remote server that agents call directly.

The design’s binding constraint is a standalone guarantee. A daemon with no farm role configured:

  • opens no new listeners or outbound connections;
  • exposes the same tools, schemas, resources and results;
  • loads no farm code at startup;
  • needs no new configuration, credentials or dependencies;
  • keeps the agent-facing MCP endpoint (including --host and its per-daemon bearer token; see Daemon);
  • leaves the Desktop App’s behaviour against a local daemon unchanged.

Background

  • Device farm design: roles, session flow, tunnel, protocol, identity, capacity, loopback, plugins, Desktop App views, build order and risks.