How novique-ai/retinue works
Retinue is a local-first application for hiring named AI teammates and coordinating them in shared rooms. The principal workflow is: a person configures retainers and a room in the web interface, sends a message, and the rooms platform adapter applies turn-taking while Hermes runs the selected agents; replies appear in the shared transcript. Room work can use a shared or isolated Podman/Docker workspace. Routines, voice, and IDE-attached workspaces are additional documented capabilities. The tree omits the documented web and gateway implementations, so those connections are shown only at the level supported by the README.
Main entry points: retinue_rooms Rooms adapter · conversation_loop.py Agent conversation loop · pipeline.py Speech humanization · engine.py Turn-taking rules
Components 12
Room experience
- Web interface
- Speech humanization
plugins/platforms/retinue_rooms/speech_humanize/pipeline.py
Room coordination
- Rooms adapter
plugins/platforms/retinue_rooms - Turn-taking rules
plugins/platforms/retinue_rooms/engine.py
Agent runtime
- Hermes gateway
- Agent conversation loop
agent/conversation_loop.py - Agent tools
agent/tool_executor.py
Workspaces and routines
- Container workspace
- Podman or Docker
- Room routines
plugins/platforms/retinue_rooms - IDE host mount
Other components
- Room participant
Connections 11
- Room participant → Web interface: uses
- Web interface → Hermes gateway: sends room requests
- Hermes gateway → Rooms adapter: dispatches room events
- Rooms adapter → Turn-taking rules: applies turn-taking
- Rooms adapter → Agent conversation loop: runs agent turns
- Agent conversation loop → Agent tools: dispatches tools
- Rooms adapter → Web interface: publishes room updates
- Rooms adapter → Speech humanization: prepares spoken text
- Container workspace → Podman or Docker: runs in
- Room routines → Container workspace: runs jobs in
- IDE host mount → Container workspace: bind-mounts