Concepts
Preview. These pages explain how the P6 runtime works and why it is shaped that way. Each states what the code does today and marks what only a spec or ADR proposes. For terms see the glossary; for commands and settings see the reference.
| Page | Question it answers |
|---|---|
| Sessions and turns | Where does a conversation live, and what survives a crash or a restart? |
| Brain and hands | What runs in the agent process, and what runs in a sandbox? |
| Session multiplexing | How do many sessions share a few warm processes, and what happens at overload? |
| Sandboxes and tiers | What isolation does a sandbox give, and how does a session find "its" sandbox? |
| Multi-user | Who can do what, and where do credentials live? |
| Dispatch archetypes | How does an orchestrator hand work to moca, and which patterns work on P6 today? |
| Context workflows | How do users move context into and out of moca, and which ways work today? |
The ideas that hold across deployments are: durable resumable sessions, a brain/hands split, several users on one deployment, and disposable sandboxes. Scale-to-zero was a property of the removed Knative/KEDA deployment, not of these ideas.