AgentPulse

A unified management dashboard for AI coding agents.
Monitor sessions, manage configurations, and track activity across Claude Code, GitHub Copilot, and more — all from one place.

---

The Problem

AI coding agents are powerful, but managing them is chaotic:

• Sessions scatter everywhere — multiple agents running simultaneously with no central view
• Context gets lost — session history, AI-generated insights, and configuration changes disappear on restart
• Configuration fragments — settings, skills, memory files, and MCP servers spread across different locations and formats
• No visibility — hard to see what agents are doing, what they've done, and what's queued

The Solution

AgentPulse is a local-first, zero-config dashboard that automatically discovers and visualizes your AI agent ecosystem. No databases, no cloud services, no setup — just install and open.

Who Is This For?

Developers who use AI coding agents (Claude Code, GitHub Copilot, etc.) daily and want a single place to monitor sessions, manage configurations, and review activity — instead of juggling scattered files, terminal windows, and config locations.

Demo

Features

Session Management

Capability	Description
Live monitoring	Real-time view of all active and historical sessions via WebSocket
Multi-agent support	Claude Code and GitHub Copilot side by side, with extensible agent architecture
Session control	Rename, close, and organize sessions from one place
Status at a glance	Active (green) vs. ended sessions, PID tracking, project grouping

AI-Powered Summaries

Capability	Description
One-click summaries	Generate AI-powered analysis of any session via claude -p
Brief naming	Auto-suggest meaningful session names from conversation content
Persistent storage	Summaries survive server restarts, cached for fast access
Title editing	Modify titles of completed sessions directly in the log

Configuration Center

Capability	Description
Global & project settings	Unified view of settings.json, .claude.json, and per-project overrides
CLAUDE.md editor	Read and edit global instructions that guide agent behavior
Memory management	Browse, edit, and delete memory files across all projects
MCP server overview	See all configured MCP servers with masked credentials
Skills inventory	Slash commands and skills organized by scope (global vs. project)

Activity & Event Log

Capability	Description
Curated Event Log	Filtered, deduplicated view of session events (oldest first)
Full Timeline	Detailed visual workflow timeline (newest first)
Smart deduplication	Consecutive identical events collapsed with count badges
Flexible filtering	Filter by type: substantive, user-only, remote input, agents, or all
Cron job tracking	View active and historical cron jobs across all sessions

Quick Start

Install from npm

npm install -g agentdashpulse
agentdashpulse

Or run without installing:

npx agentdashpulse

Install from source

git clone https://github.com/cdlliuy/AgentPulse.git
cd AgentPulse
npm install
npm start

Open http://localhost:3456 in your browser. Custom port:

PORT=8080 agentdashpulse

Run in background

PORT=3456 nohup agentdashpulse > agentdashpulse.log 2>&1 & echo $! > agentdashpulse.pid

How It Works

Zero Configuration

AgentPulse auto-discovers everything from standard locations — no config file needed:

Data	Source
Sessions	~/.claude/sessions/ and ~/.claude/projects/*/
Settings	~/.claude/settings.json and ~/.claude.json
Memory	~/.claude/projects/*/memory/
Skills	~/.claude/commands/ and .claude/commands/
Copilot	~/.copilot/

Architecture

Deliberately simple — no build step, no framework, no database:

Component	What
server.js	Express server with REST API + WebSocket — all business logic in one file
public/index.html	Vanilla JS frontend — no framework, no build step
Dependencies	Just 2: express and ws

API

22 REST endpoints + WebSocket for real-time updates:

GET  /api/sessions          — list all sessions
GET  /api/sessions/:id      — session detail with full event timeline
POST /api/sessions/:id/close — terminate an active session
POST /api/sessions/:id/ai-summary — generate AI summary
GET  /api/settings          — global + project settings
GET  /api/memory            — memory files across projects
GET  /api/projects          — all projects with configs
WS   ws://localhost:3456    — real-time session updates (5s interval)

Contributing

npm install
npm test          # 96 tests, ~85% coverage
npm run dev       # start dev server

See the contributor guide section below for code style rules, testing requirements, and review checklist.

---

For Contributors

Project Structure

agentpulse/
  server.js          — Express server, REST APIs, WebSocket, all business logic
  server.test.js     — Jest + Supertest tests (96 tests, ~85% line coverage)
  public/index.html  — Single-page frontend (vanilla JS, no framework)
  scripts/           — Screenshot & demo recording helpers (gitignored)
  package.json       — Dependencies: express, ws; devDeps: jest, supertest
  LICENSE            — MIT License

Code Style

• No build step — runs directly in Node.js, no TypeScript, no bundler
• Single-file architecture — server in server.js, frontend in index.html
• Extract shared helpers — deduplicate after 3+ occurrences
• Mask sensitive data — all MCP env configs pass through maskMcpEnv()

Testing

npm test                # all tests must pass
npx jest --coverage     # maintain >80% line coverage

Code Review Checklist

• All 96+ tests pass
• No coverage regression (>80%)
• No duplicated logic
• Sensitive data masked
• No unnecessary new dependencies

Screenshot & Demo Scripts

Helper scripts in scripts/ generate documentation assets. They require Puppeteer and a running AgentPulse server.

# Prerequisites (not included in production dependencies)
npm install --no-save puppeteer gif-encoder-2 pngjs

# Start the server first
npm start

Command	Script	Output
npm run screenshots	scripts/take-screenshots.js	4 PNGs: screenshot-sessions.png, screenshot-detail.png, screenshot-agents.png, screenshot-settings.png
npm run demo	scripts/record-demo.js	9 frames in demo-frames/ + demo.gif

Both scripts automatically mask sensitive content (file paths, usernames, project names) before capture.

License

MIT — Ying Liu