The Problem
AI coding agent sessions are siloed across editor platforms — Cursor, Windsurf, Claude Code, VS Code Copilot, and others each maintain independent conversation histories. Teams cannot aggregate token usage, compare editor effectiveness, or search across sessions without manual export and consolidation. The README states users switch between "16 editors" with "no unified view," creating operational friction for teams tracking AI investment.
What This Does
Agentlytics is a full-stack analytics dashboard built with React and Express. The ui/ directory contains a React application with 45 files including dashboard pages (Dashboard.jsx, CostAnalysis.jsx, Compare.jsx), components (KpiCard.jsx, ActivityHeatmap.jsx, ShareModal.jsx), and hooks (useLive.jsx). The server runs via server.js and mcp-server.js, likely providing the Express backend. A Deno entry point exists at mod.ts for sandboxed CLI scans. The editors/ directory contains 17 editor-integration files (cursor.js, windsurf.js, claude.js, vscode.js, zed.js, etc.) that handle session parsing for each supported platform. Configuration lives in package.json and ui/package.json (npm), with Vite build setup at ui/vite.config.js.
How To Use It
Setup: The repository uses npm as its package manager (evidenced by package.json and ui/package.json). Install dependencies with:
npm install or for the UI only: cd ui && npm install
Configuration: No environment variables or API keys are required for the core scan — the tool reads local editor data. The README notes Deno usage requires only --allow-read --allow-env permissions. For the full Node.js dashboard, no config files or secrets are documented in the repo structure.
Running it: The entry point is index.js at the root. Start the dashboard with:
npx agentlytics or directly: node index.js
The README confirms it opens at http://localhost:4637. For a lightweight CLI scan without the dashboard:
npx agentlytics --collect
Or use Deno tasks if the environment is set up:
deno task start # Full dashboard deno task scan # Lightweight CLI scan
Real-World Use
A distributed team using Cursor, Windsurf, and Claude Code can run npx agentlytics on a shared machine to generate a unified report. The dashboard surfaces KPIs (session counts, token economy), an activity heatmap (ActivityHeatmap.jsx), editor breakdown, and top models/tools. The CostAnalysis.jsx page likely visualizes spending per editor/model. Results can be exported via the ShareModal.jsx component or through the Deno --json flag for machine-readable integration with CI or billing systems.
Code Health & Issues
No test files detected — untested code paths across the React frontend and Express backend. No vitest or jest configuration is present in the file structure. No CI/CD pipeline — no .github/workflows or automated build/test gate. Pushes have no enforced quality checks. No LICENSE file — usage and redistribution rights are unclear. The README references an MIT badge, but no license file exists in the root. No input validation evidence — editor-parsing functions in editors/ directory may assume specific session formats without guardrails. Dependency hygiene — package-lock.json present, but no npm audit or vulnerability scanning configuration detected.
The Bottom Line
Agentlytics delivers a functional unified analytics dashboard for AI coding agent sessions with minimal setup — npx agentlytics launches a local dashboard at localhost:4637 that reads editor data entirely on machine. The React/Express codebase is structured and the editor-integration files in editors/ appear comprehensive for the 16 supported platforms. However, the absence of a license file, test coverage, and CI pipeline limits production readiness. It is well-suited for teams or individuals wanting quick visibility into AI token usage and editor comparison without instrumenting their own analytics, but requires manual compliance checks around licensing and testing before wider adoption.