Apple Intelligence now runs inside Screenpipe

3 min read
apple-intelligencelocal-aiscreenpipeprivacyon-device-ai

get updates on screenpipe news and releases

Apple Intelligence now runs inside Screenpipe

Update (June 2026): this feature has been removed. The Apple Intelligence integration described below — including the /ai/status and /ai/chat/completions endpoints — was removed from screenpipe in PR #4153. Builds from v2.5.40 onward do not ship these endpoints. For on-device AI, use Ollama with screenpipe instead. This post is kept for historical reference.

Your Mac already sees everything you do. Now it understands it.

Screenpipe now uses Apple Intelligence to analyze screen recordings, meeting transcripts, and browsing history on your Apple Silicon chip. This processing stays on your device.

apple intelligence × screenpipe

your screen → on-device AI → daily summary

everything processed on your Apple Silicon. nothing leaves your device.

screen capture
recording
9:15 AM🎥
Zoom

Team standup — discussed sprint priorities

capturing...
ai summary
apple silicon
waiting for data...
Processing
On-device
Data sent to cloud
Zero
Latency
< 2 seconds
Works offline
Yes

"What did I work on today?"

After a day of meetings, email, Slack, code, and research, it can be hard to remember what happened at 10am.

Screenpipe remembers everything. And now Apple Intelligence turns that raw memory into something useful:

  • Daily summaries — a structured recap of your entire day, generated in seconds
  • Action items — extracted from meetings, emails, and messages you might have missed
  • Reminders — deadlines and commitments surfaced before you forget them
  • Instant search — "what was that API endpoint?" or "what did Sarah say about the budget?"

All processed on your M-series chip. No API keys. No monthly bill. No data leaving your machine.

apple intelligencelocalhost:3030
you
what did I work on today?

How it actually works

We built Screenpipe's AI as an OpenAI-compatible API. That means any tool that works with Claude can now use Apple Intelligence instead — just point it to localhost:3030.

your app / AI client
Claude, Cursor, any OpenAI client
screenpipe server
OpenAI-compatible API (Rust/Axum)
apple intelligence
Foundation Models framework (Swift FFI)
apple silicon
M1/M2/M3/M4 neural engine

Claude Desktop, Cursor, or any AI client can use your on-device model as a drop-in replacement. Streaming, tool calling, structured JSON — it all works.

# Ask about your day — processed entirely on your Mac
curl http://localhost:3030/ai/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "apple-intelligence",
    "messages": [{"role": "user", "content": "what did I work on today?"}]
  }'

Why this changes everything

Screen data can include passwords, financial information, personal messages, medical records, and work secrets. Apple Intelligence processes that context on-chip.

It's free. No per-token pricing. No $20/month AI subscription for processing. Apple Intelligence is included with your Mac. Screenpipe is open source.

It works offline, including on a plane or with unreliable Wi-Fi.

It's fast. No network roundtrip. Responses run on-device without depending on a hosted model endpoint.

For the curious

The integration code is available in the screenpipe repository. We wrote a Swift bridge to Apple's Foundation Models framework, compiled it into Screenpipe's Rust core via FFI, and exposed it as an OpenAI-compatible endpoint. The framework is weak-linked so Screenpipe still runs on older macOS versions.

Code lives in crates/screenpipe-apple-intelligence/.

What's next

  • Windows AI — same on-device processing, coming to Windows
  • Proactive insights — AI surfaces patterns before you ask
  • Weekly reviews — auto-generated summaries of your entire week

Try it

Download Screenpipe and enable Apple Intelligence on macOS 26+ to process your screen history on-device.

get updates on screenpipe news and releases