Claude Pulse

One command. Global memory for Claude Code.

npx claude-pulse init

Claude Pulse captures every session, every decision, and every line of code across all your projects — automatically, locally, and queryable anytime. It gives Claude Code a persistent memory and gives you a dashboard to see your work.

No cloud. No accounts. No data leaves your machine.

What it does

How to set it up

What you need

1. Claude Code (install guide)
2. Node.js 18+ (nodejs.org)
3. jq and sqlite3

# macOS
brew install jq sqlite3

# Ubuntu/Debian
sudo apt install jq sqlite3

Install

Option A: Global install (recommended)

git clone https://github.com/Clemens865/Claude-Pulse.git
cd Claude-Pulse
npm install
npm run build && node bin/copy-static.mjs
npm link
claude-pulse init

Now claude-pulse works from any directory. Use Claude Code normally — everything is captured in the background.

Option B: Local install (use with npx)

git clone https://github.com/Clemens865/Claude-Pulse.git
cd Claude-Pulse
npm install
npx claude-pulse init

With a local install, prefix all commands with npx and run them from the project directory.

View your dashboard

claude-pulse start    # global install
npx claude-pulse start  # local install

Open http://localhost:3141.

The Dashboard

Overview

Key numbers at a glance: total sessions, net lines written, hours spent, active projects, streaks. Daily activity for the last 14 days. Top tools, skills, and frameworks.

Brain

A searchable timeline of your reasoning — not just your code. Structured entries for progress, decisions, blockers, patterns, and fixes. Filterable by type and project. This is the log of your thinking across sessions.

Projects

All projects ranked by activity. Click any project for session history, most-edited files, daily breakdowns.

Timeline

90-day GitHub-style heatmap. Darker squares = more active days.

Session Detail

Full event-by-event replay of any session. Every edit shows an expandable diff view. User identity (who@hostname) displayed when available.

Settings

Database info, record counts, and an export section with project/date filtering. Download your data as JSON or CSV anytime.

Commands

Terminal

Note: If you installed locally (without npm link), prefix all commands with npx.

Export supports filtering:

claude-pulse export --format json --start 2026-03-01 --end 2026-03-31
claude-pulse export --format csv --table insights

Slash commands (inside Claude Code)

These work globally — any project, any directory:

Remembering things

Some knowledge emerges mid-session. /pulse-remember captures it:

/pulse-remember fix: sqlite3 needs absolute paths on macOS
/pulse-remember pattern: all API routes use getDb() singleton
/pulse-remember context: deploy freeze next week for mobile release

Claude distills your input into a concise summary and stores both layers: - Content — Claude's clean one-liner (shown in dashboard) - Reasoning — your original words (preserved for context)

These show up as "Knowledge" at your next session start — before yesterday's progress, because accumulated knowledge is more valuable.

Audit trail

Claude Pulse captures a complete activity trail suitable for AI code audits:

• User identity — every session records who (whoami) and where (hostname)
• Diff capture — edit events store the actual old/new content, viewable in the dashboard
• Decision trail — every DECISION insight is timestamped, project-tagged, and searchable
• Export — download filtered data as JSON or CSV from the dashboard or CLI
• Integrity check — claude-pulse verify reports record counts, coverage, and SQLite integrity

For teams asking "what did the AI do?" — Claude Pulse provides every action, every decision, timestamped and queryable.

How it works

Claude Code session
       |
       v
Hook events (SessionStart, PostToolUse, Stop)
       |
       v
bash script + jq + sqlite3 → ~/.claude-pulse/tracker.db
       |
       v
Next.js dashboard (localhost:3141)

1. claude-pulse init registers three hooks in Claude Code's settings:
2. SessionStart — records session with user identity
3. PostToolUse — records each tool call with diffs (runs async, non-blocking)

4. Stop — finalizes session, computes summaries, captures structured insights

5. Smart gating: at session end, the hook counts write events. Fewer than 3? Auto-close silently. More? Ask Claude for a structured summary (PROGRESS/DECISION/BLOCKED).

6. Context injection: at session start, the hook injects last session's summary, accumulated knowledge, and cross-project awareness into Claude's context.

7. The dashboard reads the same SQLite file directly. No API keys, no cloud, no accounts.

Troubleshooting

No data showing up? Run claude-pulse doctor — it checks deps, hooks, DB, and recent activity.

jq or sqlite3 not found? Install them (brew install jq sqlite3 or apt install jq sqlite3), then re-run claude-pulse init.

Dashboard won't start? Check if port 3141 is in use: lsof -i :3141.

Start fresh? Delete the DB and re-init:

rm ~/.claude-pulse/tracker.db
npx claude-pulse init

Remove completely:

claude-pulse uninstall   # removes hooks
rm -rf ~/.claude-pulse        # removes data

Database

SQLite with WAL mode at ~/.claude-pulse/tracker.db:

For developers

Tech stack

Next.js 16 + React 19 + Tailwind CSS 4 + better-sqlite3 + TypeScript 5.9 + bash/jq/sqlite3

Development

npm run dev       # Dev server on port 3141
npm run build     # Production build
npm run lint      # ESLint

Plugin structure

The repo also works as a Claude Code plugin: - .claude-plugin/plugin.json — metadata - hooks/hooks.json — auto-registered hooks - skills/ — slash commands

Demo data

Visit Settings in the dashboard or POST to /api/seed for 30 days of sample data.

License

MIT