chore: doc update

[maxy-shpfy]

TL;DR

Introduces the Conversations & Participants model to the documentation, replacing the hardcoded "one human, one Prime" session model with a general multi-party architecture backed by a durable SQLite store.

What changed?

A new docs/overview/11-conversations-and-participants.md page defines the two core concepts: a Conversation (a transcript plus its membership) and a Participant (a session-scoped identity of kind human, agent, or automation). Existing overview pages for Sessions, Session Chat, Sub-agent Orchestration, Artifacts, Triggers, and Agent Memory are updated to reference this model — replacing language like "one agent, one thread" with the multi-party framing, introducing @mention resolution, reaction predicates, and the resource catalog.

Two new server architecture docs are added:

• docs/server/conversations.md — covers the ConversationRouter as the single entry point for all messages, the FanOutEngine and its reaction predicates (always, fromHumans, mentionsMe, atRunEnd, never), cross-conversation relay rendering as reports, transcript visibility projection, and the ResourceCatalog.
• docs/server/connectors.md — covers the ConnectorRegistry as the total resolver from participant to transport, the four connector kinds (pi-stdio, remote-env, external-inbound, a2a), connector facets replacing the old host label, the Run abstraction, and A2A point-to-point vs shared-room participation.

Existing server docs are updated throughout:

• index.md replaces the orchestrator-centric framing with the conversation-layer framing, updates the component map and dependency-injection wiring to reflect SqliteSessionStore, ConnectorRegistry, ConversationRouter, RunRegistry, ParticipantService, and the four gateways, and expands the glossary and topology model.
• agent-communication.md replaces PiAgentManager-as-hub with ConnectorRegistry-as-hub and notes that internal routes now resolve through the registry and post through the router.
• human-prime-subagents.md reframes the choreography as membership configuration over the conversation layer rather than hardcoded routing rules.
• orchestrator.md scopes PiAgentManager to the local Pi connector only.
• sessions-and-storage.md replaces the in-memory store description with the SQLite schema, documents the conversations, participants, memberships, runs, resources, resource_references, and resource_grants tables, and notes that transcripts are append-only JSONL that survive restarts.
• triggers.md updates firing delivery to go through ConversationRouter with an automation participant as author.
• ui-server-protocol.md documents per-Conversation Socket.IO rooms, new REST endpoints for participants, memberships, and resources, conversation:subscribe, participant:presence, resources:updated, and seq-ordered message delivery.
• egress-and-security.md corrects the durability note to reflect that session state now persists in SQLite rather than being in-memory.

How to test?

Read through the updated overview pages starting with 01-sessions.md and 11-conversations-and-participants.md to verify the multi-party model is coherent end-to-end. Cross-check the server docs by following the request lifecycle in index.md through conversations.md → connectors.md → orchestrator.md and confirming the component map, DI wiring, and glossary are internally consistent. Verify all internal links between documents resolve to existing files.

Why make this change?

The previous documentation described a fixed "one human talks to Prime, Prime talks to sub-agents" topology backed by an in-memory store. The architecture has moved to a general participant and conversation model with durable SQLite storage, multiple connector transports (including remote-env, external-inbound, and A2A), and server-side fan-out driven by per-membership reaction predicates. The documentation needed to reflect this so that the overview and server architecture docs accurately describe what the system actually does and give contributors a correct mental model for extending it.

[maxy-shpfy]

Warning

This pull request is not mergeable via GitHub because a downstack PR is open. Once all requirements are satisfied, merge this PR as a stack on Graphite.
Learn more

• poc: workflow debug visualization #82
• refactor: workflow layer #81
• chore: doc update #80 👈 (View in Graphite)
• feat: embed mode of ui #79
• refactor: move to unified model #78
• feat: remote agents #63 : 1 other dependent PR (#64 )
• refactor: extract ui-extensions-sdk #62
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.