Ask about images, recordings, and live sources

This guide covers Grid 0.66.8 and requires an updated app and runtime. First set up AI and select an image-capable model. Prepared local vision packages still need signing, publication, and an end-to-end inference run in Grid; they are not available merely because this guide describes the interface.

Bring a picture into the conversation

In Chat or Grid Agent, choose Add source → Images, paste an image, or drop one into the message box. Review the thumbnail, write your question, and send. PNG and JPEG are supported: up to eight images, 4 MiB each, 8 MiB combined, and four million pixels per image.

The source picker also offers:

• Grid view for the visible Grid webview on macOS desktop. It may omit native child browser or rendering surfaces, so review the captured image.
• Share window where your browser supports screen sharing. Choose a window, then Attach frame when its preview shows what you need.
• Camera after permission is granted. Choose Attach frame to take a picture.
• Connected sources to attach the latest image from a configured camera or video connector.
• Selected images from an already-loaded Data or App image input.
• Earlier in this conversation to reuse a recent attachment.

Opening a preview does not send images to AI. Stop preview releases the source; closing the picker or changing conversations also releases it.

Attach an image from a connected camera or stream

In desktop with the Pro AI runtime, open Connectors → Camera and video. Choose Images for AI under Use this source for, then choose a camera, video file path, or network stream address and start the source. This requires the installed FFmpeg decoder and any camera permissions requested by your system. Your current workbook stays open.

Return to Chat → Add source → Connected sources and choose Attach latest image. Review it, write your question, and send. You can open Connectors from this picker if you have not configured a source yet. Refresh updates the list; stopped sources are labeled as last-image snapshots.

Starting the connector does not make AI calls. Stop in Connectors ends capture. A finished recording stops without looping. The connector keeps the latest image when delivery is slow; it does not keep a recording. The existing Spreadsheet values option remains available for numerical camera output.

These images carry their arrival time, but not an original camera capture time or recording date. Use them for snapshot questions; timestamp-checked live observation requires a source that supplies known capture times.

Compare a recording

Choose Add source → Video file and open a local recording. You can seek to one moment and choose Attach frame, or use Compare a time range:

1. Enter the From and To positions using seconds, m:ss, or h:mm:ss.
2. Choose two to eight frames, within your remaining attachment allowance.
3. Review the thumbnails and their clip positions, then write your question and send.

For example: “What changed between these moments, and when did it first appear?” Sampling stays on your device until you send the selected images. This is image analysis of selected moments; it can miss events between frames and does not include sound. Keep the recording for later questions.

Observe a live source

Open Camera or Share window. Under Observe this source, enter your question and choose Start observation. Observation limits sets the capture interval and call limit. The default is one capture every five seconds and twelve calls; sessions stop within ten minutes.

Grid keeps a small window of recent images and one inference at a time. It shows capture times, rejects stale answers, and labels an answer that ages out as Earlier observation. Choose Stop observation to stop AI while keeping the preview, or Stop preview to release the source too.

Observe selected input works with an already-running connector or App input that supplies timestamped images. It does not start a connector or connect a network stream for you. Images for AI connector snapshots do not provide the capture timestamps required here; attach them through Connected sources instead.

Observation is read-only and temporary. It does not execute Agent tools, change workbook values, resume after restart, or retain original recordings. Answers and their source/time information remain in Tasks → Activity during the run's lifetime. A remote request already sent may still incur a charge after Stop.

Keep history meaningful

Conversation attachments are stored locally within the conversation limits; they are not a shared media archive. Keep original files separately when you need long-term analysis. A frame's position in a clip is not its calendar capture time.

App and connector authors can supply bounded, timestamped image observations to workbook AI, including explicit historical windows and live deadlines. The source must already provide the image data; selecting it does not start a connector, acquire a network stream, or create an archive.

Audio, direct PDF input, and native video-model input are not included. Export selected PDF pages as images when you want to ask about their visual content.