​
Inbound (you → the agent)

Send a photo, document, audio, or video in the private chat. The agent sees a local copy of the file. In a group, a photo or file is ignored unless the message also @mentions the bot or replies to it. The agent then sees the caption, not the file. See Group chats. On a fresh macOS install, local voice transcription is set up automatically. On Linux and Windows, set it up manually to have a Telegram voice note or audio file handled as message text. Telegram can only fetch 20 MB. Trim or compress a longer clip, or send it as a smaller file. A dropped connection is not that cap — send the same file again if the bot says Telegram dropped the connection. After a successful turn, those downloads are deleted.

​
Outbound (the agent → you)

The bridged agent writes files into a per-turn outbox directory (the reply names it). Anything left there when the turn ends is delivered:
  • Images (PNG/JPEG/GIF by content, not extension) → Telegram photo, unless you force a document.
  • Video (MP4, MOV, MPEG, WebM, AVI, MKV, …) → Telegram video. Non-MP4 is transcoded with local ffmpeg when possible; otherwise sent as a file.
  • Everything else → document.
Caps: 10 files / 50 MB aggregate per turn. Overflow is reported, never silently dropped.

​
Captions

If there is exactly one file and the text fits 1024 characters, text rides as the caption. Otherwise text goes first as its own message.

​
SVG

SVG is never sent as a photo (XML, and Telegram cannot render it as a photo). It goes as a document.