Channels: The Interaction Layer

Channels are the interaction layer of the product triptych (Entities, Views, Channels). They are where humans, agents, and external systems exchange messages, keep context, and drive changes.

Canonical channel model

Synap uses 4 channel types:

• thread — unified conversation type
• feed — proactive card stream
• external — synced external conversation
• agent_collab — multi-agent collaboration surface

Legacy labels (ai_thread, entity_comments, document_review, view_discussion, personal, sub_thread, external_import, a2ai) are historical terms and not canonical types.

thread: unified conversation type

thread is the default conversation container for user/AI collaboration. Instead of separate channel types, behavior is expressed via attributes:

• contextObjectType / contextObjectId (workspace, entity, document, view, project, etc.)
• threadKind: "personal" (personal assistant behavior without a separate personal type)
• parentChannelId + thread metadata for branch-style workflows (instead of sub_thread)
• agentType / routing metadata

This collapses personal and branch semantics into thread + attributes.

feed: proactive cards, actionable

feed is a read-oriented stream of proactive items (briefings, summaries, sync updates, alerts). Each card supports:

• Primary action (default continuation, open, apply, acknowledge)
• Secondary action (dismiss, snooze, open details, defer)

Feeds can exist at user or workspace scope depending on product surface.

external: connector-backed conversations

external represents conversations synced from platforms like Slack, Telegram, email, or WhatsApp.

• Imported history remains traceable to source metadata
• Reply routing back to the external platform is a capability toggle
• That toggle depends on connector live state and permissions (not all external channels are bidirectional at all times)

agent_collab: internal multi-agent coordination

agent_collab is for durable collaboration among multiple AI agents, with optional human oversight.

• Current visibility target: admin/owner only
• Intended for orchestration, handoffs, and long-running collaboration context

AI routing

Routing is policy-driven, not type-count driven:

• thread and external may route to AI depending on channel settings and agent routing config
• feed is system/proactive publishing oriented
• agent_collab routes through multi-agent orchestration logic

Context and graph linkage

All channels remain first-class knowledge graph nodes. They can link to entities, documents, views, and proposals, and they provide structured conversation memory for both humans and agents.

Migration note for docs

When older docs mention old taxonomy, map them as:

• ai_thread / entity_comments / document_review / view_discussion → thread
• personal → thread with personal attributes
• sub_thread / branch → thread with parent/branch attributes
• external_import → external
• a2ai → agent_collab

---

:::info Related pages

• Branching conversations
• Multi-agent system :::