The whole meeting, not just your half

One thing before the release notes.

The screenpipe source is moving from MIT to a source-available license: free for personal, non-commercial, educational, and research use, paid for commercial use, regardless of company size. The signed app and its subscription do not change, and personal use stays free. We wrote up the full why and exactly what it means for you here: an update to screenpipe's license.

Subscriptions are also live. Lifetime is retired — if you already have a lifetime license, nothing changes, you keep the app forever.

What's new on screenpipe this week?

The headline is meetings. Until now a recorded transcript mostly kept your side of the conversation. Now it keeps everyone, labels who said what consistently across live and recorded transcripts, and you can export the whole thing as an MP4 with synced audio. Your agents can only act on what they can see and hear — this week screenpipe captures a lot more of the room.

Please update the app, it might be outdated

Or download it

🎙️ Meetings

Consistent "who said what." Live speaker IDs now run through the same global embedding pipeline as recorded transcripts, so the same person is the same person across both.
Every participant, not just you. Recorded transcripts keep all sides of the conversation.
No more duplicate lines. Cross-device mic echo is dropped from the live transcript.
Live transcripts on the timeline, plus screen-share and calendar-based meeting detection.
Per-app ignore list for meeting detection, so the apps you never want recorded stay out.
Cloud users default to live cloud transcription.
Drag and drop images straight into meeting notes.
Meeting and date-range MP4 export with synced audio — from the CLI, the app, and MCP.
"Coming up" merges all your Google accounts into one view.

🎥 HD recording

Decoupled high-fps HD recorder — real high-fps capture, fully indexed, so you can review an important meeting or session in HD.
HD chunks now play back at native framerate in the timeline.
The "+ HD" tray click confirms with an in-app notification.

🔐 Privacy and redaction

Keyboard and clipboard capture are now OFF by default, with toggles if you want them.
Per-field PII redaction for text and image, including IDs, URLs, and dates.
Redaction happens at rest now, not at query time.
New text PII model: 266 MB down to 149 MB — faster and smaller.
New image PII model with better detection on real screens.
Text and image PII models now leverage the ANE (macOS) and NPU (Windows).
Configurable structured-ID categories.
LAN device discovery (mDNS) is opt-in, off by default.

💬 Chat

Source citations open the exact capture — straight to search or the timeline.
Per-conversation composer drafts — no more text leaking across chats.
Paste documents with attachment bubbles and an extraction spinner.
Connection chips with one-click "Try in Chat."

🔌 Connections

Redesigned connections page with a categorized layout.
New: Odoo, Krisp (one-click OAuth MCP), Plaud (OAuth MCP), Mochi, OpenClaw.
Connect multiple Obsidian vaults from the UI.
Import your device skills into the agent — cards offer both API and CLI skills.
Rotation-safe cross-device connection sync.
Agent-facing ICS calendar connection API.

⚙️ App and settings

Searchable settings.
App-wide font size setting for text scaling.
Getting started tutorial card in the Help section.
Privacy filters surface installed-but-uncaptured apps, so nothing slips through unnoticed.

🧰 CLI and SDK

CLI: work-hours schedule via --schedule-enabled and --schedule-rule.
CLI: --ignored-urls is now actually applied to capture.
SDK: native Rust telemetry for the Tauri plugin — no webview, no CSP.

⚡ Performance

Reduced timeline memory usage.
HD encode decoupled from the writer, so chunks are true constant-framerate.
Content dedup bypassed during HD sessions.
SQLite corruption guard: mmap disabled, redact worker backs off.
3× lower CPU when using more than 2 monitors.
Reduced CPU usage from the meeting detector.
Reduced overall CPU usage on Windows.

🐞 Fixes

Capture and a11y

Eliminated ghost keystrokes on macOS (removed the probe tap, graceful shutdown).
Enhanced accessibility mode is set once per app, not every 60s.
Fixed Apple OCR missing bounding boxes.
Honor the Windows OCR language selection.
Recover vision capture after a locked-screen startup, and resume capture after login.

Audio and meetings

Never drop live transcript segments on a coverage-window miss.
Fixed a CoreAudio process-tap crash.
Fixed Windows mics going silent with echo cancellation on USB mics (e.g. Logitech C922).
Keep live transcripts on the matching audio device.
Restored browser meeting detection on Windows.
No more duplicate ongoing meetings.

UI and chat

Fixed cursor placement and text selection breaking mid-sentence.
Fixed queued chat messages disappearing when switching chats.
Fixed the timeline Live Text overlay leaking over chat and blocking input.
Arrow-right timeline navigation keeps working while recording.
Chat context is now bounded in the compaction layer.

Stability

Fixed a macOS tray use-after-free crash on menu click.
Retired the top crashes (DB CANTOPEN, null model).

OAuth, platform, and pipes

Fixed the GitHub connection.
Restored the Apple Calendar permission flow.
Logout no longer needs two clicks.
Windows: signed CLI binary, no more auto-destruct false-kills, hard-fail on missing VC++ DLLs.
Scheduled pipes honor your local timezone, and missed cron slots run when the app restarts late.

Download screenpipe

Update screenpipe

Questions? Join our Discord.

Using screenpipe in a team? Let's talk

Report a bug on GitHub

Louis

screenpipe #16 | the whole meeting, not just your half