PRD: Unified Analytics Page

Status: In Progress — Frontend complete, backend & cleanup pending Branch: analytics-page Created: February 2026

---

Completion Status

DONE — Frontend Implementation

Item	Files	Status
Analytics page route & tabbed layout	app/analytics/page.tsx, components/analytics/analytics-page.tsx	Done
Overview tab (summary cards, plan usage, memory, recommendations)	analytics-overview.tsx	Done
Agents tab (summary + sortable table + health badges)	analytics-agents.tsx	Done
Workflows tab (summary + trend chart + sortable table)	analytics-workflows.tsx	Done
Documents tab (summary + RAG perf + never-accessed alerts)	analytics-documents.tsx	Done
LLM & Costs tab (tokens, cost trend, per-model, per-agent)	analytics-costs.tsx	Done
Admin tab (platform metrics + workspace table + CSV export)	analytics-admin.tsx	Done
AI recommendations (dismiss/snooze, type-based styling)	analytics-recommendations.tsx	Done
Plan usage bars (color-coded quotas, pilot-mode nulls)	analytics-plan-usage.tsx	Done
Automatos Memory card (learned vs user-set)	analytics-memory.tsx	Done
Unified analytics hooks (8 hooks aggregating existing APIs)	hooks/use-unified-analytics.ts	Done
Sidebar updated — "Analytics" visible to ALL users	components/layout/sidebar.tsx	Done
Dashboard redirects to /analytics	app/dashboard/page.tsx	Done
Removed monitoring tab from Workflows	components/workflows/workflow-management.tsx	Done
Removed performance tab from Agents	components/agents/agent-management.tsx	Done
Removed analytics tab from Documents	components/documents/document-management.tsx	Done

PENDING — Post-Merge Tasks

Item	Notes
Merge with latest main	Resolve any conflicts from recent PRs
Browser testing	Verify all 6 tabs render correctly with live backend
TypeScript verification	Run tsc once node_modules are installed
Delete old analytics files	monitoring-tab.tsx, agent-performance.tsx, analytics-tab.tsx, document-analytics.tsx, performance-analytics.tsx — only after confirming no regressions
Backend: Plan usage endpoint	GET /api/analytics/plan-usage — workspace plan limits + current usage
Backend: Admin cross-workspace endpoint	GET /api/admin/analytics/workspaces — aggregated stats across workspaces
Backend: AI recommendations endpoint	GET /api/analytics/recommendations — LLM-generated optimization suggestions
Database: New tables	workspace_plans, workspace_usage, token_usage_log (see Data Model section)
Plan tiers configuration	Define actual limits after pilot monitoring — currently using null (no limit)

---

Introduction

Replace all scattered analytics (workflow monitoring tab, agent performance tab, document analytics tab, /analytics page, /dashboard page) with a single unified /analytics page that serves two audiences:

1. Workspace Users — Actionable intelligence: "What's costing me money? How do I fix it? Are my agents/workflows/documents performing well? Is Automatos suggesting smarter choices?"

2. Super Admins — Operational monitoring: Cross-workspace costs, API usage, token consumption, plan utilization, and workspace-level reporting.

Both views live on the same /analytics route. Super admins see an additional "All Workspaces" tab with elevated data.

The Problem Today

Analytics is scattered across 5+ locations:

• /analytics — Admin-only "Intelligence & Learning" page (AI benchmarking, pattern recognition)

• /dashboard — Admin-only system health metrics

• /workflows → Monitoring tab (memory, RAG, execution stats)

• /agents → Performance tab (resource usage, task history, activity logs)

• /documents → Analytics tab (usage patterns, search performance, RAG effectiveness)

None of these are accessible to regular users. There's no cost/token visibility, no LLM optimization suggestions, and no plan/quota tracking.

What Changes

• New: Single /analytics page accessible to ALL users (not admin-only)

• Remove: Analytics tabs from /workflows, /agents, /documents pages entirely

• Remove: Existing /analytics (Intelligence & Learning) page — replaced by unified page

• Remove: /dashboard page — relevant metrics folded into unified analytics

• Keep: Individual pages (/agents, /workflows, /documents) retain their management functionality, just without analytics tabs

---

Goals

• Give workspace users clear, non-technical visibility into what they're spending and how well things are working

• Surface AI-powered recommendations (e.g. "Switch Agent X to a cheaper LLM — it's a simple email task")

• Show plan usage vs. quotas so users understand their consumption

• Provide super admins with cross-workspace cost, usage, and API monitoring

• Consolidate all analytics into one well-organized page, eliminating redundancy

• Make the page accessible to ALL users (not admin-only)

---

User Stories

US-001: Analytics page route and role-based layout

Description: As a user, I want to access /analytics from the sidebar so I can see my workspace analytics.

Acceptance Criteria:

• /analytics route accessible to ALL authenticated users (remove requiredRole: 'admin')

• Sidebar item renamed from "Intelligence & Learning" to "Analytics" with BarChart3 icon

• Page renders a tabbed layout: Overview | Agents | Workflows | Documents | LLM & Costs

• Super admins see an additional Admin tab (6th tab)

• Role detection uses existing useSystemRole() hook

• Typecheck/lint passes

• Verify in browser using dev-browser skill

US-002: Overview tab — workspace health at a glance

Description: As a user, I want a quick summary of my workspace so I know if anything needs attention without diving into details.

Acceptance Criteria:

• Top row: 4 summary cards — Total Agents (active/total), Total Workflows (active/total), Documents Stored (count + size), Monthly Cost (current period spend)

• Each card shows trend indicator (up/down arrow + percentage vs. previous period)

• Plan usage bar below cards: "Using X of Y agents", "X of Y GB storage", "X of Y API calls" with progress bars color-coded (green < 70%, yellow 70-90%, red > 90%)

• "Recommendations" section below: AI-powered suggestion cards (see US-010)

• Time range selector: 7d / 30d / 90d (default 30d)

• All data fetched from API, loading skeletons while fetching

• Typecheck/lint passes

• Verify in browser using dev-browser skill

US-003: Agents tab — per-agent performance

Description: As a user, I want to see how each of my agents is performing so I can identify underperformers and optimize them.

Acceptance Criteria:

• Agent summary cards: total agents, average success rate, total tokens used, total cost

• Agent performance table with columns: Name | Status | Success Rate | Avg Run Time | Tokens Used | Cost | LLM Model | Last Run

• Table is sortable by any column

• Agents flagged with warning badge if: success rate < 80%, cost is top 20% of agents, or using an expensive LLM for simple tasks

• Filter by: status (active/inactive)

• Clicking an agent row expands to show: execution history (last 20 runs), error breakdown, token usage over time (sparkline)

• Typecheck/lint passes

• Verify in browser using dev-browser skill

US-004: Workflows tab — workflow execution analytics

Description: As a user, I want to see how my workflows are performing so I can spot failures and bottlenecks.

Acceptance Criteria:

• Workflow summary cards: total workflows, executions this period, overall success rate, avg execution time

• Workflow performance table: Name | Runs | Success Rate | Avg Duration | Tokens Used | Cost | Last Run | Status

• Table sortable by any column

• Execution trend chart: daily runs (success vs. failure stacked bar) for selected time range

• Clicking a workflow row expands to show: recent execution timeline (last 20), step-by-step breakdown with per-step timing, error log for failed runs

• Workflows flagged if: success rate < 80%, avg duration increased > 20% vs previous period

• Typecheck/lint passes

• Verify in browser using dev-browser skill

US-005: Documents tab — storage and RAG effectiveness

Description: As a user, I want to see which documents are being used and which aren't, so I can improve my knowledge base.

Acceptance Criteria:

• Summary cards: total documents, total storage used (+ quota bar), total RAG queries this period, avg retrieval confidence

• Documents table: Name | Type | Size | Uploaded | Last Accessed | RAG Queries | Confidence Score

• Sort by any column, default sort by "Last Accessed" (never-accessed documents float to top with "Never used" badge)

• "Unused Documents" alert card: count of documents never retrieved by RAG, with suggestion to improve tags/keywords

• RAG performance section: top search queries, avg response time, retrieval success rate, sources distribution

• Documents flagged if: never accessed, low confidence scores, or no tags/keywords

• Typecheck/lint passes

• Verify in browser using dev-browser skill

US-006: LLM & Costs tab — token usage and model efficiency

Description: As a user, I want to see my LLM token usage and costs broken down by model so I can optimize spending.

Acceptance Criteria:

• Summary cards: total tokens this period (input + output), total LLM cost, cost per successful task, most expensive agent

• Cost breakdown chart: stacked area chart showing daily cost by LLM model over time range

• Token usage table: LLM Model | Requests | Input Tokens | Output Tokens | Total Cost | Avg Cost/Request | Used By (agent count)

• Per-agent cost breakdown: which agents are using which models and costing how much

• Cost trend: line chart comparing current period to previous period

• Plan quota section: "API calls: 8,423 / 10,000 this month" with progress bar

• Typecheck/lint passes

• Verify in browser using dev-browser skill

US-007: Automatos Memory section (within Overview)

Description: As a user, I want to review and tweak what Automatos remembers about me so my AI interactions are more personalized and accurate.

Acceptance Criteria:

• "What Automatos Knows" card on Overview tab showing key stored preferences

• Each memory item shows: key, value, last updated, source (user-set vs. learned)

• "Automatos learned this from your interactions" vs "You set this" visual distinction

• Edit button on each item to correct/update values inline

• Link to full memory management page if one exists, or expand to show all items

• Typecheck/lint passes

• Verify in browser using dev-browser skill

US-008: Admin tab — cross-workspace monitoring

Description: As a super admin, I want to see usage and costs across all workspaces so I can monitor the platform and identify billing concerns.

Acceptance Criteria:

• Tab only visible to users where isAdmin === true

• Top-level metrics: total workspaces, total users, total API calls (all workspaces), total token usage, total platform cost

• Workspace table: Workspace Name | Plan | Users | Agents | Workflows | API Calls | Tokens Used | Cost | Status

• Export to CSV button for workspace usage data

• Table sortable, filterable by plan tier

• Clicking a workspace row expands to show per-workspace breakdown

• Cost trend chart: stacked area by workspace over time

• "Over quota" alert section: workspaces approaching or exceeding plan limits

• API call monitoring: top API endpoints by call volume, error rates per endpoint

• Typecheck/lint passes

• Verify in browser using dev-browser skill

US-009: Remove existing scattered analytics

Description: As a developer, I need to remove analytics tabs from individual pages so users have one source of truth.

Acceptance Criteria:

• Remove monitoring-tab.tsx content from /workflows page (memory, RAG, tools tabs)

• Remove agent-performance.tsx content from /agents page (resource usage, task history, performance trends, activity logs)

• Remove analytics-tab.tsx and document-analytics.tsx from /documents page

• Remove existing /dashboard page and route (redirect to /analytics)

• Update sidebar: remove "System Dashboard" item, rename "Intelligence & Learning" to "Analytics"

• Existing management functionality on each page (create/edit/delete agents/workflows/documents) remains untouched

• No broken imports or dead code left behind (old files still on disk — delete after testing)

• Typecheck/lint passes

US-010: AI-powered optimization recommendations

Description: As a user, I want Automatos to tell me how to save money and improve performance without me having to analyze data myself.

Acceptance Criteria:

• Recommendation card format: icon + title + description + action button

• Recommendation types: cost, performance, document, quota — with type-based styling

• Each recommendation has a dismiss/snooze option

• Max 5 recommendations shown, sorted by estimated impact

• Recommendations engine that analyzes: agent LLM usage vs. task complexity, token consumption patterns, document retrieval effectiveness, workflow failure patterns (requires backend)

• Recommendations use LLM to generate natural language explanations from structured data (requires backend)

• Typecheck/lint passes

• Verify in browser using dev-browser skill

US-011: Plan usage and quota display

Description: As a user, I want to see how much of my plan I've used so I know if I'm approaching limits.

Acceptance Criteria:

• Plan usage component showing current plan name and tier

• Usage bars for each tracked resource: agents (count), workflows (count), documents (storage GB), API calls (monthly count), token usage (monthly tokens)

• Color coding: green (< 70%), yellow (70-90%), red (> 90%)

• Shows "No limit set (pilot)" when limit is null

• "Upgrade" link/button when any resource exceeds 80%

• Plan comparison tooltip or link

• Quota data fetched from real workspace plan configuration (requires backend)

• Typecheck/lint passes

• Verify in browser using dev-browser skill

---

Functional Requirements

• FR-1: The /analytics page must be accessible to ALL authenticated users (not admin-only) — DONE

• FR-2: The page must use a tabbed layout: Overview, Agents, Workflows, Documents, LLM & Costs, and (admin-only) Admin — DONE

• FR-3: All data must be fetched from backend APIs via React Query hooks — no hardcoded mock data — DONE (hooks aggregate from existing endpoints; new dedicated endpoints pending)

• FR-4: Time range selector (7d/30d/90d) must apply globally across all tabs — DONE

• FR-5: Tables must support sorting by clicking column headers — DONE

• FR-6: Plan quota bars must reflect real workspace plan limits from the backend — PENDING (using null/pilot placeholders)

• FR-7: AI recommendations must be generated by calling an LLM endpoint with structured usage data as context — PENDING (frontend cards built, backend endpoint needed)

• FR-8: Admin tab must query cross-workspace data and only render for isAdmin === true users — DONE (frontend built, backend endpoint needed)

• FR-9: All metrics must show loading skeletons (not spinners) while data fetches — DONE

• FR-10: Export functionality: Admin tab supports CSV export of workspace usage data — DONE

• FR-11: Responsive layout: works on desktop (1024px+) and tablet (768px+). Mobile is not required. — DONE

• FR-12: The page must reuse existing shadcn/ui components and Recharts for charts — DONE

• FR-13: Existing analytics components on /workflows, /agents, /documents pages must be removed entirely — DONE (imports removed; old files still on disk for safety)

• FR-14: The /dashboard route must redirect to /analytics — DONE

• FR-15: Memory preferences section must allow inline editing of stored user preferences — PENDING

• FR-16: Document analytics must surface "never accessed" documents with actionable suggestions — DONE

• FR-17: Agent analytics must flag agents using expensive LLMs for simple tasks — DONE

• FR-18: Cost data must be broken down by LLM model, by agent, and by time period — DONE

---

Non-Goals (Out of Scope)

• No billing/payment UI — No credit card forms, invoices, or payment processing. Just display usage vs. plan limits.

• No plan management — Users cannot upgrade/downgrade plans from this page (link to external billing page is fine)

• No real-time WebSocket streaming — Polling on page load + time range change is sufficient. Real-time can come later.

• No report generation — PDF/scheduled reports will be a separate PRD (Panda integration mentioned by user)

• No mobile optimization — Tablet and desktop only

• No notification system — No email/Slack alerts when quotas are hit. Display-only for now.

• No custom date ranges — Fixed presets (7d/30d/90d) only. Custom date picker is a future enhancement.

• No A/B testing of recommendations — Ship recommendations v1, measure later

---

Design Considerations

Layout Structure

/analytics
├── Header: "Analytics" + time range selector (7d/30d/90d) + Refresh + Export
├── Tab Bar: Overview | Agents | Workflows | Documents | LLM & Costs | [Admin]
│
├── Overview Tab
│   ├── 4x Summary Cards (Agents, Workflows, Documents, Monthly Cost)
│   ├── Plan Usage Bars (agents, storage, API calls, tokens)
│   ├── Automatos Memory Card (editable preferences)
│   └── AI Recommendations (up to 5 cards)
│
├── Agents Tab
│   ├── 4x Summary Cards
│   ├── Filter Bar (status, model, health)
│   └── Sortable Agent Table (expandable rows)
│
├── Workflows Tab
│   ├── 4x Summary Cards
│   ├── Execution Trend Chart (success/fail stacked bars)
│   └── Sortable Workflow Table (expandable rows)
│
├── Documents Tab
│   ├── 3x Summary Cards (docs, storage quota, RAG queries)
│   ├── "Unused Documents" Alert
│   ├── RAG Performance Section
│   └── Sortable Document Table
│
├── LLM & Costs Tab
│   ├── 4x Summary Cards (tokens, cost, cost/task, most expensive agent)
│   ├── Cost Breakdown Chart (stacked area by model)
│   ├── Token Usage Table (by model)
│   ├── Per-Agent Cost Table
│   └── Plan Quota Section
│
└── Admin Tab (super admin only)
    ├── Platform-level Metrics (5x cards)
    ├── Workspace Table (expandable rows)
    ├── Cost Trend Chart (by workspace)
    ├── Over-Quota Alerts
    └── Export CSV Button

Existing Components Reused

• Card, CardContent, CardHeader, CardTitle — from shadcn/ui

• Tabs, TabsContent, TabsList, TabsTrigger — from shadcn/ui

• Badge, Progress, Button, Select — from shadcn/ui

• Recharts — AreaChart, BarChart, LineChart, PieChart for all visualizations

• motion from framer-motion — for entry animations (match existing page patterns)

• useSystemRole() — for admin detection

• useAnalyticsSuccessRate, useAllMetrics, useCostAnalysis — existing hooks extended in use-unified-analytics.ts

Tone for Users

• Non-technical language: "Your email agent is using an expensive AI model" not "GPT-4 token consumption exceeds threshold"

• Actionable: Every metric should answer "so what?" — surface the insight, not just the number

• Friendly: Recommendations should feel like a helpful assistant, not a warning system

---

Technical Considerations

Frontend Architecture

All components live in /frontend/components/analytics/:

• analytics-page.tsx — main layout, tabs, time range state

• analytics-overview.tsx — Overview tab content

• analytics-agents.tsx — Agents tab content

• analytics-workflows.tsx — Workflows tab content

• analytics-documents.tsx — Documents tab content

• analytics-costs.tsx — LLM & Costs tab content

• analytics-admin.tsx — Admin tab content

• analytics-recommendations.tsx — AI recommendation cards

• analytics-plan-usage.tsx — Plan quota bars

• analytics-memory.tsx — Automatos memory preferences

Hooks in /frontend/hooks/use-unified-analytics.ts (8 hooks):

• useAnalyticsOverview — aggregates agent/workflow/document counts + cost

• useAgentAnalytics — per-agent performance data

• useWorkflowAnalytics — per-workflow execution data

• useDocumentAnalyticsUnified — document usage + RAG effectiveness

• useCostAnalyticsUnified — token/cost breakdown by model and agent

• usePlanUsage — plan limits vs current usage (pilot: null limits)

• useRecommendations — AI suggestions (staleTime: 24h daily cache)

• useWorkspaceMemory — stored user/agent preferences

• useAdminWorkspaceAnalytics — cross-workspace stats (admin only)

Backend API Requirements (PENDING)

New or extended endpoints needed:

Endpoint	Purpose	Status
GET /api/analytics/overview	Overview summary (agent/workflow/doc counts, period cost)	Pending
GET /api/analytics/agents	Per-agent performance table data	Pending (using existing agent APIs)
GET /api/analytics/workflows	Per-workflow execution data	Pending (using existing workflow APIs)
GET /api/analytics/documents	Document usage + RAG effectiveness	Pending (using existing doc APIs)
GET /api/analytics/costs	Token usage + cost by model, by agent	Pending
GET /api/analytics/recommendations	AI-generated optimization suggestions	Pending
GET /api/analytics/plan-usage	Current plan + usage vs. quota	Pending
GET /api/analytics/memory	User preference memory items	Pending (using existing memory APIs)
PUT /api/analytics/memory/:id	Update a memory preference	Pending
GET /api/admin/analytics/workspaces	Cross-workspace usage (admin only)	Pending
GET /api/admin/analytics/export	CSV export of workspace data (admin only)	Pending

All endpoints must accept ?days=7|30|90 and ?workspace_id=<uuid> query params.

Data Model Additions (PENDING)

-- Plan/quota tracking (new table)
CREATE TABLE workspace_plans (
  id UUID PRIMARY KEY,
  workspace_id UUID REFERENCES workspaces(id),
  plan_tier VARCHAR(50) NOT NULL,  -- 'starter', 'small_business', 'business', 'enterprise'
  max_agents INTEGER,
  max_workflows INTEGER,
  max_storage_gb FLOAT,
  max_api_calls_monthly INTEGER,
  max_tokens_monthly BIGINT,
  created_at TIMESTAMP DEFAULT NOW(),
  updated_at TIMESTAMP DEFAULT NOW()
);

-- Usage tracking (new table, updated daily)
CREATE TABLE workspace_usage (
  id UUID PRIMARY KEY,
  workspace_id UUID REFERENCES workspaces(id),
  period_start DATE NOT NULL,
  period_end DATE NOT NULL,
  agent_count INTEGER DEFAULT 0,
  workflow_count INTEGER DEFAULT 0,
  document_count INTEGER DEFAULT 0,
  storage_used_gb FLOAT DEFAULT 0,
  api_calls INTEGER DEFAULT 0,
  tokens_used BIGINT DEFAULT 0,
  total_cost FLOAT DEFAULT 0,
  created_at TIMESTAMP DEFAULT NOW(),
  updated_at TIMESTAMP DEFAULT NOW()
);

-- Token usage per model per agent (new table)
CREATE TABLE token_usage_log (
  id UUID PRIMARY KEY,
  workspace_id UUID REFERENCES workspaces(id),
  agent_id INTEGER REFERENCES agents(id),
  workflow_execution_id INTEGER REFERENCES workflow_executions(id) NULL,
  llm_model_id INTEGER REFERENCES llm_models(id),
  input_tokens INTEGER NOT NULL,
  output_tokens INTEGER NOT NULL,
  cost FLOAT NOT NULL,
  task_type VARCHAR(100),  -- for recommendation engine
  task_complexity VARCHAR(20),  -- 'simple', 'moderate', 'complex'
  created_at TIMESTAMP DEFAULT NOW()
);

Existing Infrastructure Leveraged

• LLMModel table already stores input_cost_per_1k_tokens and output_cost_per_1k_tokens

• Agent.model_usage_stats JSON field tracks total_tokens, total_cost, total_requests

• WorkflowExecution tracks started_at, completed_at, status, duration

• AgentStatistics tracks success_rate, avg_execution_time, total_executions

• Existing React Query hooks (use-analytics-api.ts, use-agent-api.ts, etc.) aggregated in unified hooks

Performance

• Tab content lazy-loads (only fetch data when tab is selected)

• Agent/workflow/document tables: paginate at 25 rows, load more on scroll

• Admin workspace table: paginate at 20 rows with server-side pagination

• Cache analytics data for 60 seconds (React Query staleTime)

• Recommendations cached for 24 hours (staleTime: 86400000)

---

Code Reuse Audit

Hooks Reused & Extended

• use-analytics-api.ts — useAllMetrics(), useCostAnalysis(), useAnalyticsSuccessRate()

• use-agent-api.ts — useAgents(), useAgentPerformance(), useAgentLogs()

• use-workflow-api.ts — useWorkflows(), useWorkflowStatsDashboard()

• use-document-api.ts — useDocumentAnalytics(), useUsageAnalytics(), useProcessingAnalytics()

• use-usage-analytics-api.ts — useUsageAnalytics() for document RAG tracking

• use-orchestration-data.ts — useAggregatedOrchestrationData()

Patterns Adapted

• Metric card pattern from monitoring-tab.tsx — glass-card with icon, value, trend

• Sortable table pattern from analytics-tab.tsx — document type breakdown

• Chart tooltip style — consistent dark tooltip from performance-analytics.tsx

• Skeleton loading from agent-performance.tsx — loading states

• Time range selector pattern — Select with 7d/30d/90d

• Badge status indicators — color-coded status badges

Design Tokens (from globals.css)

• glass-card class for all cards

• card-glow for hover effects

• gradient-text for accent headings

• Chart palette: --chart-1 through --chart-5

• Orange brand color: hsl(16 100% 55%/60%)

---

Files to Delete After Testing

These files are no longer imported but remain on disk until testing confirms no regressions:

• frontend/components/workflows/monitoring-tab.tsx

• frontend/components/agents/agent-performance.tsx

• frontend/components/documents/analytics-tab.tsx

• frontend/components/documents/document-analytics.tsx

• frontend/components/analytics/performance-analytics.tsx (old pre-unified analytics)

---

Success Metrics

• All users can access /analytics (not just admins)

• Users can identify their most expensive agent within 10 seconds of opening the page

• Users can see plan usage vs. quota without navigating away

• AI recommendations surface at least 1 actionable suggestion per workspace with active usage

• Admin can view cross-workspace costs sorted by highest spend in under 5 seconds

• Zero analytics-related components remain on /agents, /workflows, /documents pages after migration

• Page loads with data in < 2 seconds on broadband

---

Open Questions — RESOLVED

1. Plan tiers definition — TBD. Running a pilot first to monitor real usage, then setting limits. For now, use placeholder limits in the UI that can be configured later.

2. Recommendation frequency — Cache daily (not per page load). Recommendations are expensive to generate and usage patterns don't change that fast. Refresh once per day or on manual trigger.

3. Memory scope — Per workspace. Also includes agent-specific memory (e.g., communication agent remembers John's email, your Slack channel preferences).

4. Document access tracking — RAG tracking already exists via /api/documents/analytics/usage endpoint and useUsageAnalytics hook. Reuse this.

5. CSV export — Simple initial implementation matching workspace table columns. Can iterate later.

6. Dashboard removal — Remove /dashboard entirely. Audit all dashboard widgets first — move useful patterns into unified analytics. Don't rewrite from scratch where existing code works.