Record a walkthrough your AI can investigate
Use Video mode to show a bug, design, or product flow while explaining what you see. Eavesy saves the recording locally and processes signed-in audio into a transcript and brief. Share the text and the visual evidence your agent needs.
Screen, microphone, and computer audio
Video mode captures the walkthrough after you grant the required macOS permissions.
Keep the source files
The recording stays in a normal folder with available transcript, brief, and metadata files.
Choose the evidence to share
Add screenshots or a supported video attachment when your task needs visual inspection. A transcript alone does not describe every visible detail.
Clear processing boundaries
On-device captions can provide preliminary words. Sign in for the cloud transcript and brief. Version 0.5.4 does not automatically extract a visual frame timeline.
Related features
Record one complete reproduction
Open the launcher with ⌘⇧Space and choose Video. Show the starting state, repeat the action, and explain the expected and actual result. Stop after one reproduction. Verify the saved video has your voice before recording a longer session.
What the walkthrough produces
Mac 0.5.5 added walkthrough context for AI agents. Supported captures save context.md, walkthrough.json, sampled screenshots, and the original video. The preview shows context readiness; Copy supplies narration, extracted screen text, warnings, and file references.
| Output | Use it for | Keep in mind |
|---|---|---|
context.md | A readable handoff | Review readiness and warnings |
walkthrough.json | Structured context and media references | Keep the recording folder together |
| Sampled frames | Inspecting a visible state | A brief transition can fall between samples |
screen.mp4 | Original screen and audio evidence | The recipient needs a supported way to view it |
| Transcript | Checking spoken narration | Corrected full text may not be time-aligned |
OCR is extracted screen text, not a complete interpretation of the scene. An agent should inspect the relevant images when layout or visual state matters. Ask it to identify the source files and actual frame timestamps it used.
Hand context to an agent
For a local coding agent, provide the permitted capture folder or configure the read-only local walkthrough MCP server. For a browser chat, paste the context and attach the relevant screenshots or video through its supported input. Pasting a local path does not upload the referenced media.
A useful request is: “Inspect the disabled Export button after I switch workspaces. State the observation first, then investigate the condition controlling it.” The worked example provides a fictional fixture you can practice with.
Local files and cloud audio
Screen extraction and OCR run locally. Signed-in audio processing uses the authenticated transcription service. The walkthrough feature does not upload the screen video or screenshots into the hosted meeting MCP library. Sharing those files with another service is a separate action.
If you are still using 0.5.4, use the saved video, available transcript, and a manually selected screenshot. That version does not produce the newer walkthrough context. Check your installed version and the files your capture actually contains before following the local MCP setup.
Capture it once. Use the context anywhere.
Capture on your Mac and keep normal files. Sign in for cloud transcripts and briefs, then share the context your AI needs.