AI digests

A digest reads what several chats posted over the last hours, asks a model to summarise it and publishes the result to one destination on a schedule. It is the answer to "I follow twelve channels and want one morning brief".

Creating a digest

Open the Digests tab and tap New digest. The settings are split into four sections.

Sources

Add source picks a connected account and then one of its chats — a channel, a group or a private chat the account is a member of. Add as many sources as you like; a digest needs at least one to be saved. On every run the bot takes up to the last 200 messages of each source and keeps those inside the window.

Template

• News — the model groups the messages by topic and writes a digest with bullet headlines and one or two sentences each.
• Discussion — for group chats: the main threads of the conversation, who said what at a high level, and the conclusions. Greetings and small talk are skipped.
• Custom prompt — your own instruction: describe how the model should group and format the messages.

Both ready-made presets keep the language of the messages and produce plain text that fits into one Telegram message.

History depth — how many previous digests are given to the model as context (0–10) so it does not repeat what was already covered.

Schedule

• Window (hours) — how far back to read on every run; 24 by default.
• Schedule — one or more slots, each a set of weekdays and a time of day, in the time zone you pick. The digest runs at the exact minute of every slot.

Publish

• Pick destination — an account and one of its chats.
• Publish as — Bot or Connected account. The bot must be a member of the destination chat with the right to post; if the bot cannot send, the digest is published from the account instead.
• Silent — deliver without a notification sound.

Running and results

A digest is Active or Paused — the switch at the top of its settings. A run can also be started by hand: it begins within a minute regardless of the schedule and publishes for real. In the list a clock marks a digest that is generating right now, and a warning marks one whose last run failed — the error is kept with the digest; fix the cause and run it again.

Every run is logged with the number of messages read and the published text. If no source posted anything in the window, the digest says so instead of inventing content; if the model fails, a bullet list of the source messages is published so the slot is not lost.

Digests are generated on our shared key with a model chosen by us; your own OpenRouter key from AI processing is not used here.

Deleting

Delete digest removes the digest together with its run history.