OpenClaw Telegram media and attachments

Telegram has more media shapes than most channels, and each one has its own rule: an album is not ten photos, a voice note is not an audio file, a video note has no caption, and a location is neither text nor media. OpenClaw maps its portable send action onto each of them. Here is what goes out, what comes in, and the sticker path with its cache.

Albums, audio, video

• Sending several image attachments in one message tool call groups consecutive photos into Telegram albums of up to ten images in their original order, automatic replies with several photos use the same grouping, and a single photo, including a final remainder of one, is sent separately.
• The caption goes on the first photo, text that exceeds the caption limit follows the album as a separate message, and photos sent as documents, other media types and messages with inline buttons keep separate sends so their existing controls and delivery behaviour are preserved.
• Telegram distinguishes voice notes from audio files: the default is audio-file behaviour, the audio-as-voice tag in the agent reply or the as-voice flag on a send forces a voice note, and inbound voice-note transcripts are framed as machine-generated untrusted text in the agent context while mention detection still uses the raw transcript so mention-gated voice messages keep working.
• Telegram also distinguishes video files from video notes, sent with the as-video-note flag, and video notes do not support captions, so any provided message text is sent separately.

Location sends cannot be combined with message text or media.

Locations and venues

A location goes through the existing send action with one standalone location object: coordinates alone send a native pin, and adding both a name and an address sends a native venue card, with an optional accuracy field. The docs' example places the Eiffel Tower with its latitude, longitude, a twelve-metre accuracy, a name and an address. A location send cannot be combined with message text or media, so a pin and its explanation are two messages.

Stickers

• Inbound static WebP stickers are downloaded and processed with a sticker media placeholder, while animated TGS and video WebM stickers are skipped; the sticker context exposes the emoji, the set name, the file ID, the unique file ID and a cached description, and descriptions are cached in OpenClaw's SQLite plugin state to reduce repeated vision calls.
• Sticker descriptions use the configured default image model before the shared automatic image-model selection, including the provider's MiniMax image routing, the description uses one selected model and does not try the configured fallback list if it fails, a failed description is not cached, and general media analysis can still run separately with its normal fallbacks.
• Sticker actions are enabled with the sticker gate under the channel actions; the sticker action sends a sticker by file ID, and the sticker-search action searches the cached stickers by a text query with a limit.

OpenClaw on Telegram is the channel post these attachments belong to, and OpenClaw Telegram message behaviour the delivery rules and media caps underneath them.

Why the shapes matter

A Telegram user notices immediately when an album arrives as ten separate photos or a voice note arrives as a file, which is why the grouping and the voice tag exist rather than a generic upload. OpenClaw location sharing covers the cross-channel location model the pin and venue card belong to, and OpenClaw Slack media and attachments the same media rules on Slack, where files and clips behave very differently.

On Diali

Telegram is one of the channels Diali connects from the dashboard, with the runtime config generated and replaced at each release, so the album, voice and sticker rules here describe what a hosted agent can send rather than settings you maintain. Telegram on Diali is the channel on Diali and Hosted OpenClaw on Diali the assistant behind it.

• Consecutive photos become albums of ten; the caption rides on the first.
• A voice note needs the tag; a video note cannot carry a caption.
• Static stickers are described and cached; animated ones are skipped.