The short answer
AgentPlat separates collaboration, coordination and external execution into explicit responsibilities. Rooms retain shared work; Collective Runtime organizes mission activity; Mesh connects peers; safety and evidence controls constrain and explain transitions. Applications compose the adapters and policies they need.
Map responsibilities before choosing packages
Start with the question each component answers. Where do people review work? Who assigns the next task? How do independent peers exchange state? Which boundary checks permission before an external action?
The architecture overview describes a layered system built around bounded coordination, explicit authority, recovery and retained evidence. These responsibilities help prevent one convenient integration point from becoming the accidental owner of every decision.
Rooms, missions and peers have different scopes
Agent Rooms organize shared work involving participants, tasks and artifacts. The Collective Runtime composes mission planning, team formation, assignment and recovery. Agent Mesh handles coordination between independently operating runtimes.
In an illustrative research application, a Room can hold the proposal and its human review. A collective can coordinate the roles needed to produce evidence. If those roles run across independent peers, Mesh supplies the coordination layer. This is a composition choice; installing Rooms does not require every collective capability.
Keep safety and evidence across the boundaries
Follow one consequential action through the system. A plan describes intended work. An assignment identifies an eligible executor. The effect boundary must still check whether that executor may perform the specific action under current conditions.
For your integration review, draw an ownership map with four entries: the owner of the work record, the owner of assignment, the owner of the external connector and the owner of retained evidence. Check what each passes to the next and what the receiving component must independently validate.
Add organizational evolution only when needed
Agent Morphogenesis adds governed changes to operational form, such as roles, teams or agent lineage. It composes existing subsystem owners rather than replacing their authority.
A proposed specialist or team change therefore needs a path from observation to decision and then to the subsystem that can apply it. A diagram showing a model suggesting a new role is incomplete unless it also shows who may approve and install that role.
Choose adapters as operational commitments
The package catalog maps components to implementation packages. Treat that catalog as a navigation aid and verify exact exports for your chosen version. Persistence, model providers and transport are deployment decisions with their own failure modes.
For example, choosing PostgreSQL means planning migration order, backup restoration and connection limits. Choosing a model adapter means defining credentials and checkpoint support. The architecture gives those concerns explicit places; your team still configures and operates them.
Sources and further reading
Documentation reviewed . Consult the linked documentation for current implementation details.