Select text anywhere. See what the fleet knows.
Press ⌘⇧H over any selection. Atlas reads it through the macOS Accessibility API, sorts it, and shows the right context beside your pointer.
What it recognizes
Classification is pattern matching, not a model call, so it is instant and works offline. The kinds are email address, dollar amount, link, error trace, code, company, date, phone number, person, and plain text.
What each kind does
| Person, company, email, dollar amount | A briefing fanned out across Tribe, Cadence, Courier, Trellis, and Docket: who they are, upcoming meetings, recent threads, receivables, and the Docket cards assigned to you. |
|---|---|
| Error trace | Searches Chronicle logs for the trace's last line. |
| Date | Searches Cadence for events on that date. |
| Phone number | Searches Tribe contacts. |
| Link, code, plain text | Classified and shown. No fleet lookup is made. |
When there is no selection
PDFs, images, and some apps expose no selected text. Atlas then captures a 400 by 300 point region around the pointer and runs Apple's Vision text recognition on the device at its accurate setting. The image is processed in memory and is not saved.
Warm before you press
A foreground field watcher can hand the daemon the text you are typing so the Lens is resolved before you ask. Warmed results expire after five minutes.
Relationship graph
For a person or email, the Lens grows a graph of that contact's network from Tribe. Node color follows recency, size follows meeting frequency, and an amber ring marks deal health. The layout runs on the GPU. Click a contact for its details, or drag one out to open the Bar with a question about the relationship already typed.
Commitment cards
Select the text of an email and the Lens looks for promises: yours, theirs, and agreements, with a due date and dollar amount when it can find them. It offers four records together: a Slate task, a Courier confirmation draft, a Tribe note, and a Trellis draft invoice. ⌘↩ seals them, and if any fail it rolls back the rest. On the repo's 35-sentence test set the detector had no false positives and no misses, which is a small set, not a guarantee.
Questions about this tool?
Write to Renkara Media Group and a person will answer.