For AI agents

The session folder format

Exactly what Eavesy writes to disk for every session, so you (or an agent) can read it reliably.

Every session the Mac app records is written as a self-contained folder of plain files. Nothing is proprietary — it’s all Markdown, JSON, and standard audio, so any tool or agent can read it.

Folder layout

~/Eavesy/2026-06-15-design-review/
├── summary.md       # the structured brief
├── transcript.md    # the full, timestamped transcript
├── meeting.json     # machine-readable metadata
└── audio.m4a        # the recording

The folder name is the date plus a slug of the title, so sessions sort chronologically and are easy to scan. You choose the parent directory in Settings → Files.

summary.md

The human- and agent-friendly brief: a title, TL;DR, decisions, owner-tagged action items, and open questions — the same content you can copy and paste into an agent.

transcript.md

The complete transcript in Markdown. Editable, exportable, and the source of truth the summary is built from.

meeting.json

Structured metadata for programmatic use — timestamps, participants where known, and the extracted action items. Read this when you want to ingest sessions into a script or pipeline rather than read prose:

{
  "title": "Design review",
  "date": "2026-06-15T15:00:00Z",
  "durationSeconds": 1820,
  "actionItems": [
    { "task": "Ship the new docs layout", "owner": "Will" }
  ]
}
Note

Field names are descriptive and stable. If you’re building automation on top of meeting.json, treat unknown fields as additive — we add, we don’t rename.

Using it with agents

Because everything is on disk in open formats, an agent with filesystem access can read a session with no API. See Context for AI agents for patterns, and Exporting & your files for moving it elsewhere.