AI session analysis

AI analysis turns a list of 40 tabs into a short, readable sentence you can actually remember later. TabControl uses it for three things: naming sessions, categorizing them, and surfacing a brief summary plus a few insights on demand.

Every analysis returns the same shape:

• Suggested name — a short, specific label for the session. Think “Q2 SEO research — long-tail keywords” rather than “Session from Tuesday afternoon”.
• Categories — up to three from your built-in or custom category list.
• Summary — one sentence describing what the session is about.
• Insights — a few bullet-point observations (e.g. “Most tabs are product documentation from competitors”).
• Recommendations — suggested next actions (“Consider archiving — session hasn’t been restored in 30 days”).
• Confidence — a number from 0 to 1 indicating how sure the analyzer is. Low confidence on very small or very homogeneous sessions.

TabControl supports three providers. You pick one in Settings → AI Session Analysis.

Runs entirely on your Mac, using the same Foundation Models that power Apple Intelligence. Zero data leaves your device.

• Requires macOS 26 or newer with Apple Intelligence available. Earlier macOS versions fall back to a lightweight keyword analyzer that uses the macOS NaturalLanguage framework’s part-of-speech tagging.
• No API key needed.
• No rate limits.
• Works offline.

This is the right choice for most people. The fallback keyword analyzer isn’t as sharp as Foundation Models, but it’s honest — it picks up what it can from titles and domain names and labels accordingly.

Routes analysis through OpenAI’s API using the model you’ve selected (gpt-4o-mini by default). TabControl sends tab titles, URLs, and session metadata — never page content, never cookies, never login state.

• Requires an OpenAI API key. See Bring your own key.
• Output is the same shape as on-device.

Routes analysis through Anthropic’s API using Claude (default: claude-3-haiku-20240307). Same data exposure as OpenAI — tab titles, URLs, session metadata.

• Requires an Anthropic API key. See Bring your own key.

Toggle Auto-analyze in Settings to have TabControl run analysis automatically every time you save a session. The suggested name replaces the default timestamp name, and the categories get assigned automatically.

You can still rename and recategorize afterward — auto-analyze just gives you a better starting point.

If you’d rather analyze only the sessions you’re curious about:

1. Right-click the session in the popup.
2. Choose Analyze with AI.
3. Wait a moment — on-device is usually sub-second; cloud calls take 1–5 seconds depending on your connection.

The analysis shows up in the session’s detail view and is saved alongside the session, so subsequent views are instant.

If the analyzer can’t reach your chosen provider — no network for cloud, no Foundation Models for on-device — TabControl falls back gracefully:

• Cloud analyzer fails → keyword fallback runs automatically.
• On-device Foundation Models unavailable → keyword fallback runs automatically.
• Keyword fallback fails (basically never) → session keeps its existing name and has no analysis data.

Nothing ever crashes. You always get some name, even if it’s a default timestamp.

TabControl never sends page content anywhere. Analysis only ever uses tab titles, tab URLs, and your session metadata (number of windows, tab counts, timestamps, existing tags and categories).

For on-device analysis, even that stays on-device. Nothing leaves your Mac.

For cloud analysis, the payload is a small JSON document. Your API key is stored in macOS Keychain and is used to authenticate directly with OpenAI or Anthropic — TabControl doesn’t proxy through a Marsoft-owned server.

• Bring your own key — set up OpenAI or Anthropic analysis.
• Privacy and data — everything that does and doesn’t leave your Mac.