Sync Screenpipe activity to structured daily summaries. Uses Claude Code CLI for AI extraction.
npm install @screenpipe/syncSync your Screenpipe data to remote AI agents.
One-liner to permanently sync your screen memory to a remote server (e.g., Clawdbot, any SSH server). Your AI agent can then query your full history via SQLite.
Also extracts structured daily summaries: todos, goals, decisions, and AI insights.
``bashOne-liner - AI summary to stdout
bunx @screenpipe/sync
What It Extracts
| Category | Description |
|----------|-------------|
| Todos | Action items visible on screen or mentioned |
| Goals | Objectives, intentions, targets mentioned |
| Decisions | Choices made or discussed |
| Activities | Key tasks worked on, by app |
| Meetings | Calls, conversations, collaborations |
| Blockers | Problems, frustrations, obstacles |
| Insights | AI observations about work patterns |
Example Output
markdown
Daily Context - 2026-01-29
> Analyzed 480 minutes of screen activity
📱 Apps Used
- VS Code: ~180 min
- Chrome: ~120 min
- Slack: ~60 min✅ Todos Extracted
- Fix authentication bug in login.ts
- Review PR #234 for payment integration
- Send weekly update to investors🎯 Goals Mentioned
- Ship v2.9 by Friday
- Reach 50% activation rate💡 AI Insights
- Heavy context switching between Slack and VS Code (17 switches/hour)
- Deep focus block from 2-4pm on auth refactor
- Late session (after 10pm) - consider sleep impact
`Options
| Flag | Description | Default |
|------|-------------|---------|
|
| Save to directory | stdout |
| -h, --hours | Hours to analyze | 12 |
| -g, --git | Auto commit & push | false |
| -r, --remote | Sync via SSH | - |
| --json | JSON output | markdown |
| -v, --verbose | Debug output | false |Environment Variables
`bash
AI Provider (uses first available)
export ANTHROPIC_API_KEY="sk-..." # Claude (recommended)
export OPENAI_API_KEY="sk-..." # OpenAI fallback
export OLLAMA_URL="http://localhost:11434" # Local Ollama
export OLLAMA_MODEL="llama3.2" # Ollama modelScreenpipe
export SCREENPIPE_URL="http://localhost:3030" # Default
`AI Priority: Claude → OpenAI → Ollama → No AI (basic summary)
Use Cases
$3
bash
Run at end of day
bunx @screenpipe/sync --output ~/journal --hours 16
`$3
bash
Sync context to Clawdbot/Moltbot instance
bunx @screenpipe/sync --output ~/brain/context --git
Remote pulls via cron
`$3
bash
Get full week
bunx @screenpipe/sync --hours 168 --json > week.json
`$3
bash
Add to crontab
0 22 * ANTHROPIC_API_KEY=sk-... bunx @screenpipe/sync -o ~/context -g
`How It Works
$3
1. Query - Fetches OCR data from local Screenpipe API
2. Dedupe - Removes duplicate/similar screen captures
3. Extract - Claude analyzes content for structured data
4. Format - Outputs markdown or JSON
5. Sync - Optionally git pushes or SCPs to remote$3
1. Copy - Copies (your full history)
2. Sync - Uses rsync/scp to transfer to remote
3. Query - Remote can query SQLite directly`bash
On remote, query your full history:
sqlite3 ~/.screenpipe/db.sqlite "SELECT text FROM ocr_text WHERE text LIKE '%meeting%' LIMIT 10;"
``- Screenpipe running locally
- Bun runtime
- Anthropic API key for AI extraction (optional but recommended)
All processing happens locally. Screen data never leaves your machine unless you explicitly sync to a remote.
MIT - Part of the Screenpipe project.