Web Dashboard

The web dashboard is the recommended way to use Hindsight. It provides the richest experience — two-panel session inspection, 3D graph visualization, image previews, syntax-highlighted code, and real-time streaming. While the TUI is useful for quick terminal-based browsing, the dashboard unlocks the full power of the tool.

Starting the dashboard

# Start and open your browser
claude-hindsight serve --open

# Custom port
claude-hindsight serve --port 8080 --open

The dashboard is a React application embedded in the binary — no Node.js, no network requests, no build step. It runs entirely on your machine.


Dashboard Home ( / )

The home page shows aggregate statistics across all your indexed sessions:

MetricWhat it shows
SessionsTotal count of all indexed sessions
ProjectsNumber of distinct projects
Total CostEstimated USD cost across all sessions (from OTEL telemetry)
ErrorsTotal error nodes across all session transcripts
Conversations/daySparkline chart for the last 14 days
Token BreakdownInput, output, cache read, cache creation totals
Recent SessionsLast 8 sessions with first message preview
Top ToolsMost-used tools across all sessions (Bash, Read, Edit, etc.)
Cost by ModelUSD breakdown by Claude model variant

Session Detail ( /sessions/:id )

The most powerful view in the dashboard. A two-panel resizable layout for inspecting any session.

Layout

┌──────────────────────┬───────────────────────────┐
│   Execution List     │   Node Detail Panel       │
│   (left panel)       │   (right panel)           │
│                      │                           │
│   Click any row →    │   Full content, syntax    │
│   to inspect         │   highlighting, images    │
└──────────────────────┴───────────────────────────┘
  • Drag the divider between panels to resize
  • Preset buttons (30/50/70) snap to common ratios
  • Layout persists across page reloads

View modes

Toggle between two views using the LIST / GRAPH buttons:

LIST view — Flat chronological execution list. Each row shows:

  • Color-coded left border by node type
  • Type badge (user, asst, think, tool, result, task, prog)
  • Tool name badge (for tool calls: Read, Edit, Bash, etc.)
  • Summary text (truncated)
  • File chip (if the node references a file)
  • Error indicator (rose dot)
  • Timestamp

GRAPH view — Interactive 3D force-directed graph powered by Three.js + WebGL:

  • Nodes colored by type (cyan=user, green=assistant, amber=tool, etc.)
  • Node size scales by importance
  • Radial DAG layout with bloom post-processing
  • Click a node to select it — camera flies to focus
  • Drag to orbit, scroll to zoom, right-drag to pan
  • Performance-scaled: low-detail mode for 1000+ node sessions

Filter bar

Filter the execution list by node type:

ChipWhat it shows
UserHuman messages (excludes tool results)
AssistantClaude’s text responses
ToolTool calls + tool results
TaskSub-agent completion notifications
ErrorAny node with has_error: true
ThinkingExtended thinking blocks
PromptHigh-scoring prompts (score ≥ 40)

Additional controls:

  • Search — keyword search across message content, tool names, file paths
  • Sort — NEWEST ↓ (default) or OLDEST ↑
  • AUTO — auto-scroll when new SSE nodes arrive
  • Node count — shows filtered / total when filtering is active

Timeline scrubber

The thin horizontal bar below the filters shows the full session timeline:

  • Colored tick marks for each node (positioned by timestamp)
  • Token-per-turn cost bars as background fill
  • Blue playhead shows the selected node’s position
  • Click anywhere to jump to the nearest node
  • Start/end timestamps shown inline

Node detail panel

Click any node in the list to see its full content in the right panel:

Tool calls show:

  • Tool name (large, amber) with MCP server badge if applicable
  • Specialized rendering per tool type:
    • Read — file path header, line range (e.g., “lines 42–92”)
    • Edit — diff view with removed/added lines, file stats (−12 +28)
    • Write — file path + full content with syntax highlighting
    • Bash — command in bash syntax highlighting
    • MCP tools — short name + server badge (e.g., computer · claude-in-chrome)
    • Generic — labeled field display for simple inputs, JSON for complex
  • → RESULT button to jump to the corresponding result node
  • Token usage footer

Tool results show:

  • Output content with automatic language detection and syntax highlighting
  • File path header when available (from metadata)
  • Markdown rendering when content looks like markdown
  • Image preview with EXPAND lightbox for base64 screenshots
  • ← CALL button to jump back to the corresponding tool call
  • Error styling (red border) for failed results

Task notifications show:

  • Structured card with status badge (✓ COMPLETED)
  • Summary, token usage, tool count, duration
  • Task ID
  • Result rendered as markdown

Other node types:

  • Thinking — animated dots + italic text
  • Progress — message + optional percentage bar
  • User/Assistant — markdown-rendered content

Raw JSON

Every node has a RAW toggle button in the detail panel header — shows the complete JSON structure of the node.


Projects ( /projects )

Grid of all your project directories with:

  • Session count
  • Total size on disk
  • Last active time
  • Relative size bar (% of largest project)

Click a project to see its sessions.


Activity ( /activity )

Real-time feed of tool calls and session events from OTEL telemetry and Claude Code hooks.

StatSource
ToolsTotal hook tool events
AgentsSub-agent spawn events
PermsPermission check events
ErrorsGlobal error count (from transcripts)

Filter tabs: All, Tools, Agents, Lifecycle, Errors

The Errors panel on the right sidebar shows actual error events (PostToolUseFailure) with error messages and session links.

Requires hooks. Run claude-hindsight integrate --otel to install hooks. See Hooks Setup.


Prompts ( /prompts )

Shows user prompts across all sessions, scored by quality. Prompts with score ≥ 40 appear here. Useful for reviewing what you asked Claude and how effective each prompt was.


Skills ( /skills ) and Agents ( /agents )

Discovery views for Claude Code skills and agents found in your scan directories. Shows name, description, scope badges, and file paths.

Click any skill/agent to see its full definition and references.