Agents are AI teammates that handle signals for you. Each agent has one role and one area of responsibility.

There are exactly three roles.

Agents live in the Agents section and are attached to campaigns. One agent can work in several campaigns at the same time.

Analyst

Reads the messages that got through the free filters and decides: lead or not. In the prompt you describe the profile of your lead - who counts as a target and who definitely does not.

The main setting is the Score threshold (0-100). Messages scored below the threshold never reach the feed. The standard value is 50.

• Too many irrelevant signals - raise the threshold.
• Too few signals - lower the threshold: you will get more, but some will be less precise.

💡 Tip: launch a new campaign with the standard threshold, let it run for a few days, and adjust the value based on the real quality of the signals.

The analyst's work costs 1 credit per checked message.

Responder

Drafts the reply text for each lead, based on the company brain and on its own prompt.

How it replies

The responder settings include a Reply automatically switch:

• Off (the default) - the safe mode: the agent drafts a reply in every channel and you send it.
• On - the agent sends the reply itself, without your approval, in the channels where we can publish. In the other channels it still only drafts the text.

⚠️ Important: automatic sending works only in Telegram and Threads. On LinkedIn, Reddit, X (Twitter), Bluesky and Hacker News the agent always prepares a draft that you copy and publish yourself.

How the reply is delivered depends on the channel, and Telegrin shows this right on the agent card:

• By direct message - Telegram;
• By public comment - Threads;
• By copying - LinkedIn, Reddit, X (Twitter), Bluesky and Hacker News.

Reply language

The agent replies in the language of the lead's post or message - this rule outranks the prompt and the company brain. Separately, you can say what to do when the language cannot be determined (a short phrase with no clear language markers): take the language from the prompt, or use Ukrainian, Russian or English.

Generating a reply costs 1 credit.

Deliverer

Passes approved leads outside Telegrin. Create a deliverer from a ready-made template - the template already states where the leads go.

Delivering leads spends no credits.

Ready-made templates

The Agents page has a Ready-made templates gallery - agents with prompts already written for common tasks: a strict qualifier, finding contractors, finding candidates, an assistant responder, a comment generator, a Reddit responder and others.

Press Use - the agent is created with the prompt prefilled, and you can edit it freely afterwards. A template only fills in the text; it hard-codes nothing.

Your own agent

The Create agent button on the Agents page. Pick a type, optionally use a template as a starting point or start From scratch, give it a name and write the prompt.

Next to the prompt field there is an Improve button. It shows the suggested version separately, without changing your text - then you decide: Accept, Edit or Keep original.

ℹ️ Note: the prompt does not need to repeat the company description, the product and the tone of voice - the agent takes all of that from the company brain. The prompt answers a different question: how exactly to behave in this role.

What works with what

The analyst and the responder read the company brain; the deliverer does not. The deliverer has no prompt: it writes nothing, it only passes things on.

A campaign can be launched without agents at all - then it runs on keywords alone. That is the cheapest mode: in Telegram it is free.

Details on attaching agents to a campaign: the "Campaigns" section, article "Step 3: agents in a campaign".