Your Own Prompt Instead of a Summary: «Essence» Bot Now Processes Posts with Your AI Key

The standard summary in «Essence» Bot works for many channels, but not for all of them. For one channel you only need the numbers and dates, for another a three-point summary, and in your own channel you want posts published in its usual format. Until now, you could only choose the summary language and a regular or cynical style.

Now you can replace the standard summary with your own prompt. You connect your own API key for an AI service, choose a model and write an instruction, and the bot processes new posts according to it. Requests to the AI service are paid from your account with that service. Below is what this changes, how to set it up and where the feature has its limits. The text describes the bot as of October 12, 2026.

What changes

The bot takes the original post and, instead of the standard summary, gives the model your instruction. The result arrives where the summary used to arrive: in your private chat with the bot and in your channels and groups with auto-posting. This works for both public and private channels.

The instruction can fit any task, for example:

• “Keep only the numbers, dates and company names, as a list.”
• “Summarize in three points and explain difficult terms.”
• “Write briefly, in our channel’s format: a one-line headline, then two lines of substance, no emoji.”

The model processes the text of the post, including short posts, polls and posts with links. If a post has no text, there is no processing. The response language comes from the bot’s settings, so you don’t need to write “answer in English” in the instruction. The regular and cynical styles are not added to your own instruction: you set the tone.

The model responds the way it understood the instruction, and it sometimes makes mistakes. A good example does not guarantee that the next posts will be processed just as well.

What you need to connect it

The feature is available in the trial period, Basic and Pro. «Essence» Bot charges nothing extra for it, and your plan limits stay the same. You pay the chosen AI service for each request.

In the bot, open “Settings” → “Custom AI processing” and connect a service:

1. Choose a service: OpenAI, Anthropic Claude, xAI Grok, DeepSeek, Google Gemini (Google AI Studio) or OpenRouter. Through OpenRouter, models from different developers are available.
2. Send your API key for that service. The bot will check it and store it encrypted.
3. Choose a model. The first screen shows up to ten models; the rest can be found by name, developer or exact ID. The list contains only text models available with your key.

You can connect one service, and it is used for all your instructions. The key can be replaced: the new one is checked first, and the previous connection is replaced only after that. If you disconnect the service, the key is deleted, but your instructions are kept. Only you change the model; the bot never swaps it on its own.

How much it costs

The price of requests is set by the AI service whose key you use. «Essence» Bot has no money budget of its own for this feature. Spending and spending limits are configured in the service’s dashboard, and that is where you should set a cap if you need one.

All requests are paid: processing new posts, previews and retries after an error. If the service does not respond in time, the request may still be charged: that is up to the service, not the bot.

Where to set the instruction

The instruction can be set at three levels:

• For all posts — in the “Custom AI processing” section. Here you choose between two options: “My instruction” or “Essence summary”.
• For a single source channel — with the “Configure sources” button in the same section.
• For an auto-posting destination — with the “Custom AI processing” button on the screen of your channel or group in auto-posting.

A source channel and an auto-posting destination have three modes: “Use general setting”, “My instruction” and “Essence summary”. The last one brings back the standard summary even if a general instruction is set.

Instructions do not add up: one applies to each post. The bot looks for it in order: first the source channel setting, then the auto-posting destination, then the general one. If nothing is set anywhere, the standard summary arrives. The settings screen shows which instruction currently applies and which level it comes from.

An example. The general instruction is “summarize in three points”. For a news channel where only the numbers matter, its own is set: “keep only the numbers and dates”. For your public channel with auto-posting, “Essence summary” is selected. Then all channels arrive in your private chat as three points, and the news channel as numbers and dates. Your public channel receives standard summaries, except for the news channel: its own instruction takes precedence over the destination setting and applies both in your private chat and in all destinations. Such exceptions are visible in the destination settings.

How to connect the channel or group itself for auto-posting is described in the article about auto-posting.

Check the instruction before it goes live

The “Preview” button sends an example of processing to your private chat, with the instruction, language and formatting that will actually apply. Each new example is a request to the model through your key. If you open a ready example again, the model is not called again.

This is especially important for auto-posting: there is no approval step before publishing in your channel or group, and the result of the instruction goes there right away.

A changed instruction applies only to new posts. Results already delivered are not redone and are not published again.

Limits

• The instruction is up to 2,000 characters, the result up to 1,000 characters.
• The instruction changes how a post looks, but does not decide whether it arrives. The bot cannot select posts by topic. To skip posts by words, there are source filters.
• The model works only with the text of the post: it does not search for additional data on the internet and does not perform actions.
• If the key does not work, or the model refused or returned a result that cannot be published, processing stops and the bot sends a notification. Neither the standard summary nor the original post will arrive instead of the result, and the bot will not switch to another service or model.
• From private channels, only processed text arrives: without the original, attachments or a fragment of the source text, even on error.
• The text of the instruction may end up in the model’s response. Don’t write anything in it that the readers of your channel or group shouldn’t see.
• Links and images the model adds to the response are not published. The bot adds the source link and the signature itself.
• Group and channel digests follow their own rules; the instruction does not apply to them.

What happens to your data

The text of the post and your instruction are sent to the chosen AI service. Whether it stores them and for how long is determined by its terms, not by «Essence» Bot. The key is stored encrypted and never ends up in logs or in the request to the model itself. Results of processing with your instruction are stored separately from standard summaries and are not given to other users.

Where to start

Choose the channel whose standard summary suits you least. Connect your key in “Settings” → “Custom AI processing”, set an instruction for that channel only and request a preview. If you like the result, keep it, and later decide whether you need an instruction for the other channels and for your auto-posting destinations. If you don’t use the bot yet, add this channel first: the 5-day trial period starts with the first channel you add, and your own instruction is already available in it.