Unclassified
Workbench Master Plan
The trilogy: Projection Doctrine + Builder Doctrine + Tool Absorption Doctrine, executed in sequence.
The Three Doctrines
| Doctrine | File | Governs |
|---|---|---|
| Workbench & Projection | docs/tech/workbench-projection-doctrine.md | How relevance is computed and rendered |
| Workbench Builder | docs/tech/workbench-builder-doctrine.md | How the surface must be built |
| Tool Absorption | docs/tech/tool-absorption-doctrine.md | How external capabilities are internalized |
Workbench Surface Boundary Decision
| Workbench | Canonical Surface | Open WebUI Role | Boundary Rule |
|---|---|---|---|
| User Workbench | Projection-native customer/product shell | No shell role | The customer-facing workbench remains the IntegrateWise product shell. It may launch or hand off context into Twin, but is not replaced by Open WebUI. |
| Twin Workbench | Open WebUI primary Twin surface | Direct runtime | Open WebUI is the primary cognition surface for Twin: chat, models, prompts, knowledge, tools, automations, and memory. |
| Governance | Embedded at every layer — approval queue and HITL review surface inline in the Operational Workbench | Governance-adjacent | Approval panels appear inline, not in a separate shell. |
Canonical styling rule:
- Twin / Open WebUI runtime surfaces use Forest + Paper.
- Midnight Executive remains investor-deck language only.
Execution Sequence
Phase 1: Stabilize Projection Architecture
Goal: The Projection Engine is the foundation. It must be solid before any surface work continues.
| Task | File / Location | Status |
|---|---|---|
useProjection() hook reads all spine config tables | lib/projection/use-projection.ts | ✅ Built |
| Module resolver computes work/org/personal views | lib/projection/module-resolver.ts | ✅ Built |
| Field policy enforcement | lib/projection/field-policy.ts | ✅ Built |
| Module guard in ContentRouter | lib/projection/module-guard.ts + content-router.tsx | ✅ Wired |
| Role passthrough AppShell → WorkspaceShell → ContentRouter | AppShell.tsx → workspace-shell-new.tsx → animated-content-router.tsx | ✅ Wired |
RBAC role name fix — DB enum practitioner/readonly → FE member/viewer | Migration or mapping layer | 🔄 Needed |
ProjectedModules passthrough — useProjection() → sidebar navigation filter | workspace-shell-new.tsx nav generation | 🔄 Needed |
Tool Absorption (Phase 1):
- PostgreSQL — truth layer fully utilized via direct Spine DB RPC reads
- Infisical — secrets governance remains backend-only; no surface change yet
Phase 2: Make Navigation Projection-Native
Goal: The sidebar and navigation must be driven by projection, not hardcoded maps.
| Task | File / Location | Status |
|---|---|---|
Dynamic module resolver replaces static DOMAIN_CONTENT_MAP | content-router.tsx | ✅ Guard + module-resolver.ts dynamic resolution active |
Sidebar nav filters by tenant_products × role × department | workspace-shell-new.tsx via nav-builder.ts | ✅ buildNavFromProjection() drives sidebar |
Navigation respects depth_matrix | nav-builder.ts | ✅ Cross-dept modules grouped with depth badges |
| Product entitlement gates navigation items | module-guard.ts + nav-builder.ts | ✅ canAccessModule() gates every sidebar item |
| Three doors enforced: My Department / My Organization / My Self | Tabs in workbench + nav sections | ✅ Built |
Tool Absorption (Phase 2):
- BookStack — doctrine browser link in Org View (integrated reference stage) ✅
Phase 3: Build SDAI-Native Operational Surfaces
Goal: The Workbench must make the Schema-Driven AI Integration pipeline visible and operable.
| Task | Primitive | Backend Need |
|---|---|---|
| Ingestion Panel — real-time source connector status | Health Ring + Timeline Stream | sync_jobs polling endpoint |
| Source Connector Tiles — per-tool status (Salesforce, HubSpot, Gmail, Slack, ChatGPT, Claude, Kimi) | Entity Card variant | Connector status API |
| Schema Extraction Badge — shows which template/schema was fetched | Metric Tile | spine_schema_registry per-entity |
| Normalization Progress — 8-stage pipeline visualization | Progress bars + Timeline Stream | Per-entity pipeline state |
| Curation Queue — AI-proposed entities awaiting human judgment | Governance Panel | continuity_objects with proposed state |
| Schema Mapping Preview — tool field → spine field mapping | Operational Canvas | Schema mapping API |
New Components to Build:
components/workbench/panels/ingestion-panel.tsxcomponents/workbench/panels/connector-tile.tsxcomponents/workbench/panels/curation-card.tsx
Tool Absorption (Phase 3):
- Spine DB conversational_memory — continuity timelines visible in Timeline Stream
- Agent Zero — execution cards in Action Tiles (native-assisted stage)
Phase 4: Surface Ingestion + Curation + Write-Back Visibly
Goal: Close the loop. User sees raw input → AI curation → approval → write-back → confirmation.
| Task | Primitive | Backend Need |
|---|---|---|
| Raw Input Preview — show scrambled source data (email thread, doc, AI convo) | Operational Canvas | Source content fetch API |
| AI Curation Card — curated entity proposal with confidence | Insight Block + Entity Card | continuity_objects content |
| Schema Mapping Approval — approve field mappings before entity creation | Governance Panel | Schema mapping + approval API |
| Write-Back Proposal — "Update Salesforce stage? [Confirm]" | Action Tile | connector.execute with preview |
| Write-Back Confirmation — timeline showing write-back success/failure | Timeline Stream | workflow_executions + audit_log |
| Sync Conflict Detection — alert when source tool has diverged | Insight Block | pipeline_quarantine + diff API |
| Empty Dashboard Killer — ingestion always produces candidates | All primitives | Continuous ingestion cron |
New Components to Build:
components/workbench/panels/curation-card.tsxcomponents/workbench/panels/writeback-monitor.tsxcomponents/workbench/panels/sync-conflict-alert.tsx
Tool Absorption (Phase 4):
- n8n — workflow trace in Timeline Stream, execution in Action Tiles (native-assisted stage)
- Open WebUI — contextual Twin handoff panels linked to entities (Twin-primary, not customer shell)
Phase 5: Absorb High-Value Tool Affordances
Goal: Progressive absorption of the strongest capabilities from the external stack.
| Tool | Affordance to Absorb | Workbench Primitive | Stage |
|---|---|---|---|
| Affine | Block-based operational canvases, freeform drafting | Operational Canvas (enhanced) | Native-assisted |
| BookStack | Hierarchical doctrine navigation, canon browsing | Org View canon panels | Integrated reference |
| Open WebUI | Twin entrypoint, entity-linked chat, contextual handoff into Open WebUI | Twin launch surfaces, Insight Blocks | Twin-primary / product-non-shell |
| n8n | Workflow trace visibility, execution provenance | Timeline Stream, Action Tiles | Native-assisted |
| Beszel | Runtime health cards, system anomalies | Health Ring, Health Canvas | Integrated reference |
| LiteLLM | Model provenance, routing transparency | Governance Panel metadata | Integrated reference |
| Ollama | Local-vs-cloud reasoning visibility | Twin provenance badges | Integrated reference |
| Agent Zero | Visible execution cards, bounded task runs | Action Tiles, execution summaries | Native-assisted |
| Spine DB memory.* | Continuity timelines, memory-linked entities | Timeline Stream, memory attachments | Native-primary |
New Components to Build:
components/workbench/panels/twin-chat-panel.tsx(Open WebUI handoff, not shell replacement)components/workbench/panels/doctrine-browser.tsx(BookStack absorption)components/workbench/panels/workflow-trace-panel.tsx(n8n absorption)components/workbench/panels/execution-summary-card.tsx(Agent Zero absorption)
Phase 6: Use v0 for Final Experiential Coherence
Goal: v0 is the polish benchmark. Apply its interaction patterns to the rebuilt surface.
| v0 Pattern | v1 Location | Action |
|---|---|---|
| Working memory tiles — context + sources + next step | _deprecated/os-home-view-wired.tsx | Extract pattern, apply to EntityCard |
| Unified surface — one place, no tab switching | workspace-shell-new.tsx | Ensure Personal/Work/Org are fluid, not jarring |
Contextual action buttons — canAct flag | _deprecated/os-home-view-wired.tsx | Apply to ActionTile |
| AI memory — "You discussed pricing on Jan 28" | _deprecated/os-home-view-wired.tsx | Apply to TimelineStream |
| Low cognitive switching | All views | Audit: can user complete task without leaving workbench? |
| Visible next steps | EntityCard + InsightBlock | Every card must have "What next?" |
Builder Acceptance Test (apply to every view):
- Can a user understand what matters in under 10 seconds?
- Can a user act without opening 4 more tabs?
- Can a user see evidence before approving?
- Can a user remain inside one contextual surface for a full task cycle?
Phase 7: Expand Departments Through Configuration Only
Goal: No bespoke UI for new departments. Only configuration.
| Department | Config (entityTypes + metrics) | Workbench |
|---|---|---|
| Business Ops | ✅ Built | ✅ BizOpsWorkbench |
| Customer Success | ✅ Configured | ✅ CSWorkbench |
| Sales | opportunity, lead, quote, sequence | DepartmentWorkbench wrapper |
| Marketing | campaign, lead, email, social_post | DepartmentWorkbench wrapper |
| RevOps | pipeline, forecast, quota, cohort | DepartmentWorkbench wrapper |
| Finance | invoice, budget, expense, report | DepartmentWorkbench wrapper |
| Service | ticket, sla, knowledge_article, satisfaction | DepartmentWorkbench wrapper |
| Procurement | vendor, purchase_order, contract, renewal | DepartmentWorkbench wrapper |
| IT Admin | system, incident, vulnerability, certificate | DepartmentWorkbench wrapper |
| Product/Eng | feature, bug, sprint, release | DepartmentWorkbench wrapper |
| Student/Teacher | student, course, assignment, grade | DepartmentWorkbench wrapper |
Each new department requires only:
- Entity type list
- Metric definitions
- Thin wrapper component (20 lines)
- Module registry entry
No custom pages. No custom CSS. No custom data fetching.
Capability Adoption Tracker
| Tool | Current Stage | Target Stage | Phase |
|---|---|---|---|
| PostgreSQL | Native-primary | Native-primary | 1 (complete) |
| Spine DB memory.* | Integrated reference | Native-primary | 3-4 |
| Open WebUI | External Twin surface | Twin-primary / product-non-shell | 4-5 |
| n8n | External only | Native-assisted | 4-5 |
| Affine | External only | Native-assisted | 5 |
| BookStack | External only | Integrated reference | 2-5 |
| Beszel | External only | Integrated reference | 5 |
| LiteLLM | External only | Integrated reference | 5 |
| Ollama | External only | Integrated reference | 5 |
| Agent Zero | External only | Native-assisted | 4-5 |
| Infisical | External only | External only | Retained as specialist |
Decision Log
| Date | Decision |
|---|---|
| 2026-05-17 | Three doctrines formalized: Projection, Builder, Tool Absorption |
| 2026-05-17 | 7-phase execution sequence established |
| 2026-05-17 | SDAI (Schema-Driven AI Integration) defined as core product pattern |
| 2026-05-17 | v0 remains polish benchmark, v1 remains substrate |
| 2026-05-17 | Tool absorption is staged, not immediate replacement |
| 2026-05-17 | Department expansion is configuration-only after Phase 7 |
| 2026-05-21 | Open WebUI reaffirmed as the primary Twin surface, not the customer-facing product shell |
This plan is a living document. Update it when phases complete or architecture shifts.