AI Tools Analyzer — Node.js

Unified dashboard for every AI conversation you've ever had, running locally in one command.

Scans your local AI conversation history across six tools and renders a live, auto-refreshing dashboard with cost breakdowns, tools/models/project analytics, and a built-in Claude agent server:

• Claude Desktop (Cowork) + Claude Code CLI
• Gemini CLI
• Antigravity IDE + CLI
• ChatGPT exports
• Codex CLI
• Puku CLI

Plus a built-in MyAgent server: an Anthropic-API-powered CLI agent with prompt caching and tool execution.

This is the Node.js rewrite of the original Python claude_analyzer.py + myagent.py. Same data, same dashboard, but:

• Serves a live dashboard on http://localhost:4310 (auto-refreshing via SSE)
• Exposes all data as JSON APIs (/api/kpis, /api/sessions, /api/cost, …)
• File watcher re-scans source folders whenever you start a new conversation
• Can also emit a self-contained snapshot HTML (--snapshot)

Quick start

Run with npx (no clone required)

npx ailense                      # starts on http://localhost:4310
npx ailense --port 8080          # custom port
npx ailense --snapshot report.html  # write a self-contained HTML and exit

npx downloads the package + its dependencies into a temp cache on first run, then executes the ailense bin. Subsequent runs use the cache.

Or install globally

npm install -g ailense
ailense                          # same flags as above

Or clone and run from source

git clone https://github.com/Soab42/agent-report.git
cd agent-report
npm install
cp .env.example .env             # add your ANTHROPIC_API_KEY (optional)
npm start                        # or: node src/index.js

Then open http://localhost:4310.

For a single offline HTML file (no server):

ailense --snapshot report.html
xdg-open report.html             # or just open in any browser

Where to find things

| Thing | Where | |---|---| | Source | https://github.com/Soab42/agent-report | | npm package | https://www.npmjs.com/package/ailense | | Issues | https://github.com/Soab42/agent-report/issues | | License | MIT |

What it scans

The parser auto-detects these paths on your machine:

| Source | Default paths | |---|---| | Claude Desktop | ~/.config/Claude/claude-code-sessions/, ~/.config/Claude/local-agent-mode-sessions/ | | Claude Code CLI | ~/.claude/projects/ (fallback ~/.claude/) | | Gemini CLI | ~/.gemini/tmp/*/chats/session-*.jsonl | | Antigravity IDE + CLI | ~/.gemini/antigravity-{ide,cli}/brain/<convId>/.system_generated/logs/transcript.jsonl | | ChatGPT export | ~/Downloads/conversations.json, ~/Desktop/conversations.json, ~/Documents/conversations.json (or pass --chatgpt <file>) | | Codex CLI | ~/.codex/sessions/<YYYY>/<MM>/<DD>/rollout-*.jsonl | | Puku CLI | ~/.puku-cli/projects/**/*.jsonl | | Memory files | *.md with type: user|feedback|project|reference frontmatter under Claude/Codex roots | | Settings | settings.json, settings.local.json, .claude.json |

Disable a source with --no-claude, --no-gemini, --no-antigravity, --no-codex, or --no-puku.

CLI flags

ailense                                # start server on $PORT or 4310
ailense --port 8080                    # custom port
ailense --snapshot r.html              # write self-contained HTML report, exit
ailense --no-watch                     # disable file watcher
ailense --verbose                      # verbose source listing
ailense --path /extra/root             # add extra Claude root to scan + watch
ailense --chatgpt FILE                 # path to a ChatGPT conversations.json
ailense --no-claude                    # disable a specific source
                                        # (--no-gemini, --no-antigravity, --no-codex, --no-puku)

HTTP API

All endpoints accept ?source=<claude|gemini|antigravity|chatgpt|codex|puku> and ?period=<today|week|month|all>.

| Endpoint | Description | |---|---| | GET / | Dashboard HTML | | GET /api/scan | Per-source scan info (events, sessions, path) | | GET /api/kpis | Aggregate KPIs | | GET /api/sessions?limit=&offset= | Paginated session list | | GET /api/sessions/:id | Full session detail | | GET /api/events | Raw event list | | GET /api/cost | Cost breakdown by model and day | | GET /api/projects | Project activity | | GET /api/top | Top tools/models/projects | | GET /api/activity | Timeline + hourly + weekday | | GET /api/memory | Memory files | | GET /api/settings | Settings files | | GET /api/stream | Server-Sent Events (live updates) | | POST /api/agent | {task, model?, useTools?} → run MyAgent once | | POST /api/agent/stream | Same, with SSE stream |

Example

curl localhost:4310/api/kpis?source=claude&period=month | jq
curl -X POST localhost:4310/api/agent -H 'Content-Type: application/json' \
     -d '{"task":"list the files in /tmp"}'

Pricing

Hard-coded first-party Anthropic rates (USD per 1M tokens). Edit src/pricing.js and public/app.js (search RATES) to customize. Cache reads = 0.1× input, cache writes = 1.25× input (5-min TTL). Non-Claude tools are excluded from cost (they log characters, not priced tokens).

How it works

src/
├── index.js              # Express server + file watcher + snapshot mode
├── parsers/              # one parser per source (claude.js, gemini.js, …)
├── routes/api.js         # JSON API endpoints
├── routes/agent.js       # POST /api/agent
├── routes/html.js        # static dashboard
├── agent/                # MyAgent — Claude SDK + tools + JSONL logger
├── schema.js             # normalized event/session schema (mirrors ESRC/SSRC)
├── pricing.js            # RATES table + rateFor()/eventCost()
└── utils.js              # parseTs, extractText, walkFiles, fmt, …

public/
├── index.html            # dashboard markup
├── styles.css            # extracted from build_html <style>
├── app.js                # fetch + render + tab/filter logic
└── chart.umd.min.js      # bundled Chart.js 4.4.1

The server maintains an in-memory state = {events, sessions, memory, settings, scan} and re-scans affected roots whenever chokidar reports a file change. The browser fetches the JSON over /api/* and the dashboard re-renders — no full page reloads.

Requirements

• Node.js ≥ 18
• Optional: ANTHROPIC_API_KEY in .env to use /api/agent (the dashboard itself works without it)

License

MIT