Capture / Capturing
Capturing
A capture is an image with your laser marks on it, plus an optional intent and note. You pick a source, draw on it, check the preview, then choose who gets it.
Pick a source
| Source | Where | How |
|---|---|---|
| Camera | Android app, web app | Point the camera, then draw on the live picture. Your first stroke freezes the frame. |
| Choose an image | Android app, web app | Any image file or photo on your device. |
| Share to lwis | Android app | Share an image from another app to lwis (up to 20 MB per image, 10 at a time). |
| Capture screen | Web app on desktop browsers | Pick a window, tab or screen. Only offered where the browser supports screen sharing, so not on Android or iPhone. |
| Laser on a page | Firefox extension | Press Alt+L (or use the toolbar popup) and draw right on the web page. |
| Pick element | Firefox extension | Alt+click an element on the page. |
| Select area | Everywhere, once the image is taken or chosen | Drag a rectangle instead of drawing. |
Draw with the laser
Draw with your finger or mouse. The laser leaves a short fading trail, and strokes you make within about two seconds of each other form one zone: when you stop for two seconds, lwis moves on to the preview.
lwis recognises what you drew and tells the agent: a tap, circle, arrow, double arrow (spread), line, check mark, cross, strike-through, zigzag, scribble, area, or a circle followed by an arrow ("move this there"). The laser marks are burned into the image, and the agent also gets a description of each gesture and a cropped close-up of what you marked.
Select area is the alternative to drawing, including from the keyboard: arrow keys move the rectangle, Shift+arrow keys resize it, and Enter confirms.
Check the preview and redact
Nothing leaves your device before the preview. There you can:
- Redact: paint over anything private with a black brush. Redactions are burned into every part of the capture (the annotated image, the plain image and the close-ups) before it's sealed. Undo removes the last stroke.
- Review page text: a capture from the extension also carries the page's title, address and the text of the elements you marked. Each piece is listed with a checkbox: untick any you don't want to send.
Images are re-encoded before sealing, which drops their metadata (EXIF location, camera model and the like), and scaled to at most 2048 px on the long edge.
Add an intent and a note
The poll asks "What should the agent do?" and suggests two or three intents from your gesture. For example, a cross suggests "Remove this" and "This is wrong: why?", and an arrow suggests "Move A to B".
- When an agent session is selected, tapping an intent sends right away; otherwise it just sets the intent for the other send buttons.
- Other… lists every intent: Fix this, Explain this, Change this to…, Remove this, This is wrong: why?, Move A to B, Connect A → B, Make A look like B, Use this value, Implement what's written here, This is right, keep it, and Other.
- Add a note for anything else, such as "this one?".
- Send without intent is fine too.
Choose where it goes
On the send screen you pick one destination:
- An agent session on a paired machine: the list shows the sessions you have attached, then your hosted agents, with their project and whether they're idle or busy. lwis remembers the last session you used.
- A contact or a group, once you have an identity: see Contacts and following and Groups.
- Your own devices, such as a browser linked to your phone: see Your devices.
- A share link, with no pairing at all: a one-time link for any agent with a shell, and Copy image for chat apps. See Share links.
On Android, sending to an agent session needs a fingerprint or screen-lock check within the last 5 minutes. Sending to contacts or your devices doesn't.
An agent can also ask you for a capture: the app then shows "project asks for a capture" with its request above the sources.
Delivery states
After you send, the app shows where the capture is: Sealed, Uploaded, Sent, waiting for the agent, then Delivered. If the machine is offline you see Queued: the machine is offline, and if something fails the capture is kept in the outbox with the reason, for example "pairing expired: pair again" or "declined in the terminal".
History keeps your last 50 captures on the device for 7 days, with any replies.