--- title: "Agent" id: "1821" type: "page" slug: "agent" published_at: "2026-09-20T22:09:00+00:00" modified_at: "2026-09-21T01:36:29+00:00" url: "https://xedant.com/agents/telegram/docs/agent" markdown_url: "https://xedant.com/agents/telegram/docs/agent.md" excerpt: "The Agent section is the integration with Xedant Agent: inside Telegram Agent a chat with…" --- # Agent [https://xedant.com/agents/telegram/docs/agent.md](https://xedant.com/agents/telegram/docs/agent.md) The Agent section is the integration with Xedant Agent: inside Telegram Agent a chat with an assistant appears, one you can task with “take care of the channel”, “go through the queue”, “look at the sources” — in plain language. Without setup all chat surfaces stay hidden, and the rest of the product works as usual. ## Setup Two environment variables at startup are enough: `AGENT_API_URL` — the address of your Xedant Agent — and `AGENT_API_KEY` — its API key. After that the Call Agent buttons and the chat panel appear. How to set the variables is in the [Installation](/agents/telegram/install) section. ## Where to call it Call Agent buttons sit on the channel page, in the channel’s queue and in commenting projects. The context — the channel’s title and @name — is inserted into the prompt automatically, so the agent does not have to guess which channel you mean. ## Chats The chat list with a filter: search by title or by label, plus a choice of “With labels”, “Without labels” or a specific label. Each chat’s row shows the label, the last update time, the processing duration, the message count, the messages-per-second and tokens-per-second rate, the cost, the file-edit count, the environment and the skill. New chats appear in the list by themselves, in real time: an active chat shows “processing” and its counters, a working one a stop button. The Chats button at the bottom of the sidebar opens the same list with the same filter — no need to leave other sections. **Chat labels** are your own marks so a needed conversation does not get lost among hundreds: a label has a name, an icon and a color. A label is set right in the open chat’s header and in the list row, changes at once in every open tab, and the chat does not “jump” up the list because of it. Labels are managed in “Manage chat labels”: labels are created, renamed and deleted there; on deletion the application tells you how many chats will be left without the label. ## The chat panel The chat opens as a panel at the bottom of the screen. The agent’s messages are in Markdown; tool calls are compact cards: file edits with diffs, task lists; the current task has a pinned panel. Messages waiting in a working chat’s queue show as one-line rows above the input box (paused ones in a warning color): they can be edited, paused or removed. Right in the panel are the total token and cost counters, and a sent message appears on screen immediately, without waiting for the server’s reply. ## Voice input A task for the agent can be dictated instead of typed: the input box has a mic button — it appears when speech recognition is enabled in the settings of your Xedant Agent. The button works both in the chat panel and in the prompts queue rows. The recognized speech is inserted into the box right where you were typing — fix it up and send. Details in the [Voice Input](/agents/telegram/docs/speech) section. ## Files in messages The input box accepts clipboard contents directly — with the paste button on the computer or the Ctrl+V shortcut. A screenshot, picture or file is uploaded to the project folder, and the message text gets the path to it: the agent knows which file you mean and can open or edit it. Text from the clipboard pastes where the cursor is. On the phone the paste button is replaced by a paperclip: it opens the file picker. A file that is too large or of the wrong type will not upload — the application says what is wrong. This saves you when the document you need sits on your computer instead of in the project: no need to move it over by hand first. ## Prompts and auto-send The Prompts button at the bottom of the sidebar opens the shared task list: rows of “text + environment + skill” that can be prepared in advance. Tasks go out one by one — the bottom ones first, that is, the oldest; a sent row disappears from the list. It is a way to give the agent a series of tasks and go do other things. Automatic sending can be enabled — and the queue moves by itself: the application launches tasks one after another while you are busy elsewhere. The auto-send settings define how many chats work at once and how often to check the queue; the same place holds the fine print for those who queue tasks in batches: whether to save project changes before sending a task, whether to pause auto-fix until the queue is done, whether to send during peak-load hours, and whether to switch to the launched chat. Empty rows are not sent — they are deleted — and when the queue runs out, auto-send turns itself off. ## Skills and environments When creating a chat you can pick a skill from the agent’s list — a set of instructions for specialized work — and an environment (the model). The choice is remembered and becomes the default for the following chats. The skill is honestly shown on the chat it was created with. ## Sounds and the background While a chat works in the background, sounds arrive: a new message, an error, completion. A summary of active chats — with a timer and the cost — lives on the home page and the channel pages: it can be collapsed into a single button with the totals. Quiet mode turns the sounds off. ## License Sending messages and answering the agent’s questions requires a valid license — without one the application offers to enter a key (voice input and file pasting run into the same restriction — they are also sends into the chat). The key is activated in the license dialog, and it shows which products the key covers. Everything else in the product — channels, publishing, the schedule, the queue, sources, comments, analytics — works without a license. The important advantage: the chat is the only thing closed off by the license. Details on the [Licensing](/agents/telegram/license) page. ## The external API (v1) Every section of the product is available to scripts and integrations too: channels, bots, accounts, sources, the queue, messages, authors, comments, the schedule and analytics. API keys are created in the settings — the full key is shown once, afterwards only its beginning; a key can be revoked at any moment and stops working immediately. Interactive documentation with all the commands opens from a link in the settings and lets you try requests right in the browser. API errors arrive with a code, a clear message and a hint about what to fix. ## What is next About Xedant Agent itself — its capabilities and installation — read the [Xedant Agent](/agents/agent) section on this site. How to deploy Telegram Agent together with the integration is in the [Installation](/agents/telegram/install) section. [← Back to the documentation index](/agents/telegram/docs)