Roadmap

What's built, what's in progress, and what's next.

Last updated: March 2026

---

Completed ✅

Foundation (2025 H1)

• Event sourcing: immutable events table, pg-boss job queue, full audit log
• Hub Protocol v1: standard API contract between Data Pod and Intelligence Services
• Multi-agent core: Orchestrator pattern, 15+ specialized agents
• tRPC 11 + Drizzle ORM: end-to-end type safety throughout

Entity & Profile System (2025 H2)

• Unified entity model: one type for tasks, notes, contacts, events, ideas
• Profile system: entity types as configuration (system, workspace, shared, user scopes)
• Property definitions: typed, constrained, profile-scoped, slugged
• Knowledge graph: typed bidirectional relationships between any entities
• Profile defaultValues: property defaults merged at entity creation time
• Profile inheritance: webinar extends event

Views & Visualization (2025 H2–2026 H1)

• 8 view types: Table, Kanban, List, Grid, Calendar, Timeline, Graph, Whiteboard
• Bento dashboard: composable grid of widgets, JSONB config, react-grid-layout, edit mode
• 14+ built-in widgets: welcome, entity list, recent, quick capture, stats, view runners
• View-runner widgets: Kanban, Table, List, Calendar, Timeline adapters as bento blocks
• Entity bento mode: toggle any entity to a dashboard view; auto-creates default layout
• Widget error boundaries: crashed widget shows error card; rest of grid continues
• Unknown widget placeholder: marketplace CTA for unregistered widget types
• Field system (@synap/entity-card): single rendering source of truth across all view types

Channels (2026 Q1)

• Channel types: ai_thread, branch, entity_comments, document_review, view_discussion
• AI streaming: SSE with chunk, step, branch_decision, complete, error events
• Branch management: fork a conversation, explore alternatives, visual tree navigation
• Context items: link entities or documents to a channel for AI grounding
• Backend refactor: chat_threads → channels, conversation_messages → messages

AI Governance & Agent Users (2026 Q1)

• Proposal flow: AI executes → checkPermissionOrPropose → user approves in inbox
• Agent user system: agents are rows in users table with workspace RBAC roles
• Hub Protocol branding: source: "intelligence" / isHubProtocol: true — cannot be spoofed
• autoApproveFor[]: glob-pattern whitelist for low-stakes auto-approval
• sourceMessageId flow: proposals linked to the message that triggered them
• Security hardening: see synap-backend/docs/SECURITY_AND_PERMISSIONS.md

Template Engine (2026 Q1)

• 6 built-in templates: CRM, Second Brain, Content OS, Project Management, Personal, Blank
• Template → workspace: profiles + property defs + views + home dashboard + sidebar items
• Shared profile scoping: templates share profiles across workspaces without slug collisions
• Browser sidebar items: templates configure layoutConfig.sidebarItems for Electron app
• Home bento dashboard: every template provisions a bento view with homeScope: workspace

Intelligence Ecosystem (2026 Q1)

• ZeroClaw integration: one-click provision, Hub Protocol connection, status polling
• @synap/agent-service: generic useAgentServices(), ServiceChip, ServiceCard components
• OpenClaw architecture: Hub-and-spoke boundary defined, synap-os SKILL.md spec written
• A2AI channel type: spec for agent-to-agent async communication

Data Model Consolidation (2026 Q1)

• Removed documents.entityId backlink: reverse lookup now uses FK on entities.documentId
• Canvas-only document creation: structured views and bento never create a document row
• Profile defaultValues field: merged at entity creation, profile < caller priority

Browser (Electron) App (2026 Q1)

• Template-driven sidebar: workspace sidebar items configured by template
• View resolution: lazy view ID lookup on click
• Connection management: Data Pod connection indicator, health status
• Workspace identity chip: colored initials pill for multi-workspace visibility

---

In Progress 🚧

MCP Integration

• MCP client in Intelligence Hub: connect to external MCP servers (Playwright, filesystem, Postgres, Slack, WhatsApp)
• MCP server exposure: expose Synap data as an MCP server for Claude Desktop, Cursor, ChatGPT
• One wire unlocks 10,000+ community tools; one server makes Synap accessible from any MCP client

External Channel Relay

• Inbound webhook receiver: normalize messages from Telegram, Slack, WhatsApp, Discord
• Outbound relay: Synap AI response → Hub Protocol → external platform delivery
• Message normalization layer: uniform schema for cross-platform messages
• Priority order: Telegram (free, fast) → Slack (Bolt SDK) → WhatsApp (Meta approval, 4–8 weeks)

Background Agents

• Agents that run asynchronously, submit proposals, and surface results when the user returns
• Powered by the existing proposal queue — no new governance infrastructure needed
• Use cases: overnight research runs, daily digest generation, automated tagging

---

Next 📋

Widget Marketplace

• Registry of community-built widgets installable to any bento layout
• Manifest format: widget key, package, config schema, preview screenshot
• Security review process for marketplace submissions
• "Search widget marketplace" CTA already wired in UnknownWidgetPlaceholder

MCP Widget Type

• mcp:* widget prefix → route render to connected MCP server
• MCP server can expose a custom React component or a data feed
• Unlocks: GitHub PR status widget, Linear issues widget, Notion block embed, etc.

OpenClaw Integration

• Phase 1: Install synap-os SKILL.md → OpenClaw connects via Hub Protocol (1–2d)
• Phase 2: A2AI channel type for async agent-to-agent messaging (2–3d)
• Phase 3: External channel bridge — inbound webhooks → EXTERNAL_IMPORT → relay (3–4d)
• Phase 4: Filesystem governance (user-scoped paths auto-approve, system paths propose)

AI-Generated Widgets

• "Generate this widget" CTA in UnknownWidgetPlaceholder
• User describes what they want → AI generates a widget spec → bento block inserted
• Requires MCP + Intelligence Hub round-trip

View Subscriptions & Notifications

• Subscribe to a view's query: get notified when matching entities appear
• Use case: "notify me when a deal moves to 'Closed Won'"
• Powered by the pg-boss job queue + existing webhook dispatcher

Conflict Detection

• When multiple agents (or users) propose changes to the same entity simultaneously
• Diff-based merge suggestions in the approval queue
• Deferred until background agents are in production use

---

Deferred (Not Now)

Feature	Reason deferred
Extension marketplace (code plugins)	MCP covers the use case more cleanly
Built-in WhatsApp integration	Meta approval takes 4–8 weeks; Telegram first
Managed OpenClaw hosting	Governance unclear (creator joined OpenAI); deprioritized
Native mobile app	Electron browser covers desktop; mobile is a separate investment
LangGraph orchestration	Our multi-agent system is sufficient; LangGraph adds complexity without clear benefit

---

Design Principles for Prioritization

1. Governance first: any new AI capability must fit the proposal model before shipping
2. MCP before custom: if an MCP server exists for a tool, use it — don't build a native integration
3. Templates before settings: a good default is worth more than a configuration option
4. Entities over documents: content stored in entity relationships is queryable; content stored in documents is opaque
5. Widget error isolation: a broken widget must never break the dashboard