# Connecting to your agent Source: https://aitextura.mintlify.app/en/agents/connect Connecting Telegram and other messengers to an autonomous agent: the bot is created without a token, and only the people you specify can message the agent. An agent is only as useful as it is handy: it's faster to fire off a task from your phone on the go than to open the workspace. The **Connect** tab connects messengers where you chat with the agent just like with a person. The Connect tab: Telegram, Discord, Slack, Mattermost, IRC, Nostr, and WhatsApp cards with the “Not connected” status | Messenger | Notes | | - | - | | **Telegram** | The main option. You can create the bot right from the platform; no token needed | | **Discord** · **Slack** · **Mattermost** | For teams that already work there | | **IRC** · **Nostr** | For those who know why they need them | | **WhatsApp** | Connects the same way | These are the agent's channels, not [AI Employee channels](/en/channels). Here it's you who does the talking, not your customers, and they're configured separately for each agent. ## What you'll need * A Telegram account — on your phone or in the desktop app. * Your Telegram ID, if you're connecting the bot with your own token. To find it, message `@userinfobot` — it replies with a number. * Nothing else: no server, domain, or platform token required. ## Connect Telegram Click the **Telegram** card — a panel with two tabs opens. The Telegram panel with the “Create via us” and “Own token” tabs The platform opens `t.me`. Your browser asks for permission to open the Telegram app — allow it. Telegram opens the **Create bot** service screen: **BOT NAME** is how it's labeled in conversations, and **USERNAME** is a unique handle that must end in `bot`. Tap **Create**. A chat with the new bot opens. Be sure to tap **Start** (the `/start` command) — without it, the platform can't reach the bot. Use this if you've already created a bot with `@BotFather`. The “Own token” form: “Bot token” and “Your Telegram ID” fields and the “Insert my ID” button The **Bot token** field takes a string like `123456789:AA...` from `@BotFather`. The token goes straight into the agent's settings, isn't stored in the conversation, and is never sent back to the browser. The **Your Telegram ID** field takes numeric IDs separated by commas. The **Insert my ID** button fills in yours automatically. Find other people's IDs with the same `@userinfobot`. The Telegram card status changes to **Connected**. ## Who can message the agent The agent replies only to people whose Telegram IDs are listed in the setting. Anyone else who finds the bot gets silence. This is the only barrier between your agent and outsiders. The agent has access to your files, email, and command line — don't add the IDs of people you wouldn't trust with all of that access. Learn more in [Security and Access](/en/agents/security). To add a colleague, get their ID via `@userinfobot` and append it, separated by a comma, in the same field. ## Check it Send the bot "Hi". A fresh agent replies with an introduction — asks how to address you and suggests a name for itself. An agent that's already set up replies to the point. If you haven't gone through the [first run](/en/agents/first-run) yet, now is the time: whatever you tell it about yourself, it writes to its files. ## Troubleshooting | Symptom | Cause and fix | | - | - | | The bot doesn't reply to anything | **Start** wasn't tapped in the chat with the bot. Open the chat and tap it | | The bot doesn't reply only to a colleague | Their Telegram ID isn't in the **Your Telegram ID** field | | "Invalid token" | The token was copied with an extra space or was already revoked in `@BotFather` — generate a new one | | The bot replies, but for the wrong agent | One bot works with one agent. Create a separate bot for your second agent | | Replies arrive minutes late | The container was restarted. Check the agent's status and the [Log](/en/agents/maintenance) | ## FAQ Yes. The **Chat** button in the workspace opens the same conversation with the agent. Telegram is for your phone: that's where you'll give tasks on the go and receive [scheduled summaries](/en/agents/schedule). Message `@userinfobot` — it replies with a number. It's not your `@username` but the actual digits. You don't have to look up your own ID: the **Insert my ID** button fills it in. Add their IDs separated by commas. Each person messages the agent individually; a shared group chat is a different scenario, and for customer groups there's the [AI Employee](/en/channels/telegram-groups). Open the messenger card on the **Connect** tab and clear the token field. An empty field disconnects it. Discord, Slack, Mattermost, IRC, Nostr, and WhatsApp connect the same way: card → connection panel → access credentials. The "who can message" restriction works there too. ## What's next Your first conversation with the agent in the connected messenger Summaries that will arrive in this same chat Who should and shouldn't get access to the agent A separate bot for each of the agent's personas # Agent Files Source: https://aitextura.mintlify.app/en/agents/files A look inside the agent's working environment: the container folder tree, the file editor, and sensitive-data markers. What's safe to touch and what's better left closed. The agent doesn't work in the chat — it works on its own disk: it downloads exports, stores results, and keeps notes. The **Files** tab shows this entire environment — the tree on the left, the selected file's contents on the right, with the option to edit and save. The Files tab: the agent's container folder tree and an empty preview area Above the list are three icons: create a file, create a folder, and refresh the tree. ## What lives where | Folder | What's inside | | - | - | | `workspace` | The working folder: whatever the agent is working on for your tasks | | `identity` | Character and memory files — the same ones shown on the [Personality](/en/agents/personality) tab | | `plugins` · `plugin-skills` · `skill-workshop` | [Skills and plugins](/en/agents/skills): installed ones and the ones the agent builds for itself | | `logs` | The technical log, also known as the [Log](/en/agents/maintenance) | | `agents` · `devices` · `state` · `npm` | Container system folders | Configuration files sit in the root. A ⚠ icon next to a name marks sensitive data — keys and connection settings. Don't edit files marked ⚠ — `.env` and `openclaw.json` — by hand, and don't show them in screenshots: they contain access keys. If something breaks in them, don't fix it manually — use **Settings → Management → Diagnostics**. The **Show backup copies** toggle at the bottom of the list reveals technical backup files. It's off by default, and that's the right setting: you don't need them in everyday work. The same tree later: plugin-skills, plugins, and skill-workshop folders have appeared The tree changes as the agent works: install a skill or enable the task board, and new folders appear. That's normal. ## Why you'd open this tab Most of the time, you won't need to: the agent manages the files. Open the tab in four cases. | Situation | What to do | | - | - | | Get a result | Find the file in `workspace` and download it | | Give the agent materials | Create a file or folder and paste in the content | | See what the agent did | Check which files appeared and what's in them | | Correct its notes | Open a document in `identity`, edit it, and save | Asking the agent to handle files is usually faster than doing it by hand: "put the July report in workspace/reports and show me the result." The **Files** tab is for checking, not for daily work. ## FAQ They shouldn't: the container disk survives restarts. Make sure you're not looking in the wrong folder — the agent may have saved the result to a subfolder in `workspace`. If the file really is gone, look for traces in the [Log](/en/agents/maintenance). Yes — create a file with the icon in the panel and paste in the content. The second way is easier: send the file to the agent in chat and ask it to put it in the right folder. Give it the full path: `workspace/reports/july.csv`. The agent doesn't scan the disk continuously — it goes where you send it. They're deleted with it and can't be recovered. Before deleting, download a [backup](/en/agents/maintenance) — an archive with the agent's entire state. The interface has no separate usage counter. Rule of thumb: the container is a working environment, not storage. Keep large archives on your side, and give the agent only what it's working with right now. ## What's next A folder that instantly becomes a live website The same files, presented for easy editing Backup, diagnostics, terminal Why you shouldn't touch files with keys # First Run Source: https://aitextura.mintlify.app/en/agents/first-run Your first conversation with the agent is an introduction and briefing, following a script the platform put in its file. What it asks, how to answer, and why the first task takes half an hour. A fresh agent doesn't wait for commands — it starts the introduction itself and draws out what it needs to do its job. Half an hour for the first task and a minute for the same task a week later isn't an exaggeration, it's the usual curve: the agent writes down in its files how to do it. ## The first conversation Type "Hi" — in the workspace chat, or in Telegram if the channel is already connected. The agent's reply to “Hi”: a greeting, a question about how to address you, three name options, and a question about your routine This isn't a canned greeting. The agent's memory has a `memory/ONBOARDING.md` file — a first-run script that it follows step by step and then deletes at the end: the file's absence is the "introduction complete" marker. You can view or rewrite the script on the [Personality](/en/agents/personality) tab. The ONBOARDING.md document: the “Introduction” and “Briefing” steps with the questions the agent asks on first run ## What it asks | Step | Question | What happens to the answer | | - | - | - | | **Introduction** | How to address you: your name, and whether to be formal or casual | Saved to the **About User** document | | **Introduction** | What to call the agent — it suggests 2–3 name options | Saved to the **Personality** document | | **Briefing** | What routine eats up your time every week | Becomes the basis for the first tasks | The questions come one at a time — answer the same way, one at a time. An agent that gets everything dumped on it at once will simply lose some of it. Tell it your time zone in the very first conversation. Without it, scheduled jobs will run at the wrong time, and "send the summary at 6:00 PM" will come as a surprise. ## What to tell it right away The more precisely you describe your situation, the fewer questions it will ask later. * What your business does and who your customers are — in one paragraph. * Where the materials it will work with are: drive, spreadsheets, email. * How you want to receive results: briefly in Telegram, as a file, as a spreadsheet. * What it must not do without your permission — message people, publish, spend money. Duplicate that last point in the **Rules** document on the [Personality](/en/agents/personality) tab: the agent rereads what's written in the file in every new conversation, while what you said in the chat can be lost when the history gets compressed. ## The first task Start with the routine you named in the briefing. Here's what happens next: It asks about the format, the data source, and what counts as a finished result. That's normal: the agent isn't a genius, it's a doer that needs boundaries. Need a drive, email, or a spreadsheet? Ask right in the chat: "how do I connect Google Drive?". It explains step by step what to click and installs the missing [skill](/en/agents/skills) itself. Check the actual result, not the report — the agent may say "done" after doing half. Learn more: [Where the agent breaks](/en/agents/limits). The workflow goes into its files. Next time the same task takes minutes, because it doesn't need to figure it out again. For the first tasks, use "show me what you're going to do" mode — and only then let it act on its own. Permissions grow as trust grows; they aren't granted in advance: [Security and Access](/en/agents/security). ## FAQ That means the first-run script has already been completed and the file deleted — this happens if you chatted with the agent earlier. No big deal: just tell it about yourself in a regular message, and it will add that to its files. Yes. Everything it wrote down is stored as text on the [Personality](/en/agents/personality) tab: the **About User** document is about you, **Personality** is about the agent. Open it, edit, and click **Save**. Changes take effect in the next conversation; no restart needed. Check whether the **Memory** ability is enabled under **Settings → Abilities**: without it, the agent doesn't retain facts between conversations. If it's enabled, write the recurring fact into the **Long-Term Memory** document yourself — that's more reliable than waiting for the agent to save it. Stop it and break it into parts. A sign the task is too big: the agent goes in circles clarifying details instead of showing an intermediate result. Give it a piece, accept it, give it the next one. Open the **Personality** tab, find the `ONBOARDING.md` document in the **Memory** block, and paste the script back in — the agent will run it in the next conversation. You can easily write the script in your own words: it's plain text with steps. ## What's next The files where the agent records you and itself A board that shows what it has taken on The first scheduled job — a morning summary What to allow, and what to hold back until trust is built # Autonomous Agents Source: https://aitextura.mintlify.app/en/agents/index An agent does the work on its own: sorts your email, keeps records, compiles scheduled reports, and sends the results to Telegram. How it differs from an AI Employee and how it works. Some work isn't about talking to customers — it's routine: sort the inbox, tally expenses from receipts, put together a summary by Friday, publish a post. An **autonomous agent** does this on its own. It has its own workspace with files, memory, and a command line, and you give it tasks in plain words in Telegram. The difference from a chatbot is simple: a navigator shows you the way, a taxi driver takes you there. An AI Employee answers customers; an agent completes a task and sends you the result. ## AI Employee or agent These are two separate products on the platform. You choose on the very first step of the hiring wizard. | | [AI Employee](/en/guides/ai-employees) | Autonomous agent | | - | - | - | | Who it talks to | Your customers | You | | Where it works | In channels: website, messengers, CRM | In its own workspace; replies arrive in Telegram | | What it does | Answers from the Knowledge Base, qualifies leads, books | Processes files, browses websites, writes scripts, keeps records, publishes | | Runs on a schedule | No, it responds to incoming messages | Yes, it wakes up on its own and runs scheduled jobs | | Remembers you | Remembers the customer in their [profile](/en/guides/profiles) | Keeps memory files about you and its work | | Ready to work | In a couple of minutes | The container starts in a minute; fine-tuning takes the first day | One doesn't rule out the other: an account can have both AI Employees and agents at the same time. Both are listed in the **Agents** section of the left menu. ## How it works Four things that keep everything else from looking like magic. | Part | What it is | Where to configure it | | - | - | - | | **Personality** | Plain text files: how the agent talks, what matters to it, what it knows about you. No black box — open them and rewrite | [Personality](/en/agents/personality) | | **Skills** | Also text: an instruction file that teaches the agent something new once it reads it. Installed from the catalog | [Abilities and Skills](/en/agents/skills) | | **Heartbeat** | An internal timer. The agent wakes up on its own, rereads its instructions, and runs background jobs — no command from you needed | [Schedule](/en/agents/schedule) | | **Memory** | Working memory is finite. A long conversation gets compressed and details get lost — so one agent does one big job, not everything at once | [Personality and Memory](/en/agents/personality) | One big job — one agent. If you have a single agent track sales, book restaurants, and write code, its memory will overflow and it will start ignoring its own rules. ## The first day An agent needs some fine-tuning, and that's a normal part of the launch, not a malfunction. * The first conversation is an introduction: the agent asks how to address you and suggests a name for itself. It then writes your answers to its files on its own. * The first real task can take half an hour: it clarifies details, asks follow-up questions, and asks for help with access. * You don't need to be a programmer. Need a drive, email, or a spreadsheet? Ask in the chat "how do I connect it" — it explains step by step and installs the missing tools itself. * The same task tomorrow takes a minute: it wrote down how to do it and follows its own notes next time. ## Limits Upfront, before you get started. | Limit | What it means | | - | - | | **Two agents at a time** | You can't deploy more than two agents on an account. To free up a slot, delete one in the **Agents** section | | **The agent fails silently** | It may report a task as done when part of it isn't. Verify important results yourself — [Where the agent breaks](/en/agents/limits) | | **Access is a risk** | The agent has a command line and your connections. Grant the minimum permissions — [Security and Access](/en/agents/security) | | **Usage is billed per model** | The agent thinks with a model and spends your account balance on it. You can connect [your own provider key](/en/agents/models) | ## FAQ Who it talks to and what it's allowed to do. An AI Employee talks to your customers using the Knowledge Base and doesn't go anywhere beyond its connected channels and talents. An agent talks to you, has its own workspace with files and a command line, and completes tasks rather than answering questions. If the job sounds like "answer customers," you need an [AI Employee](/en/guides/ai-employees). If it's "do this for me," you need an agent. No. All communication is a regular chat in Telegram or in the dashboard. The agent picks and installs its own tools, and if it lacks access, it explains step by step what to click. In a container on the platform side: its own disk, its own command line, and its own public address for [sites](/en/agents/sites). You don't need to set up a server, and nothing is installed on your computer. About a minute — the container starts, the environment is prepared, and the tools are connected. You can close the page; the platform will email you when the agent is ready. Technically it can write to messengers, but that's not its job: it has no Knowledge Base, Self-Learning, Working Hours, or answer quality control. For customers, use an [AI Employee](/en/guides/ai-employees). ## What's next Five areas and real examples — where people usually start A three-step wizard and deployment in a minute What the agent asks in the first conversation and how to answer Five rules that remove almost all the pain # Where the Agent Fails Source: https://aitextura.mintlify.app/en/agents/limits An honest list of what an autonomous agent does poorly: it forgets, reports success without results, and stays silent about failed tasks. And how to catch it. Everything in the other sections holds true as long as you know about these four failures. None of them can be fixed with a setting: they're built into the nature of an agent, and you protect yourself with habits. The general principle: **check the result, not the report**. An agent is a diligent intern with no ill intent but a confident tone. ## It forgets Working memory is finite. When a conversation gets long, the system quietly condenses it, and details disappear. The agent doesn't tell you it forgot: it just carries on with whatever is left. What it looks like: the agent mixes up names and numbers, asks again about things you agreed on an hour ago, and stops following its own rules. | What to do | How | | - | - | | Keep important things in files | Permanent facts go in **Long-Term Memory**, restrictions go in **Rules**. See [Personality and Memory](/en/agents/personality) | | Don't pile everything into one conversation | New topic, new conversation | | Don't pile everything onto one agent | One big task, one agent or a separate [Persona](/en/agents/team) | | Give it a model with a larger context | The **Brain** slot in [Models](/en/agents/models) | ## It lies An agent can cheerfully report that a task is done when part of the work isn't. This isn't deception, it's a quirk: the agent reconstructs a picture of what happened and sometimes gets it wrong in its own favor. What it looks like: "logged all seven workouts," but the database has three. "Sent the emails," but only one went out. | What to do | How | | - | - | | Ask for proof, not a status | "Send me a line-by-line list of what you logged," not "done" | | Check the result yourself | A file in [Files](/en/agents/files), a row in a spreadsheet, an email in the Sent folder | | Give tasks with a verifiable outcome | Not "sort out the expenses" but "send a table with the total for each category" | ## It goes silent The most expensive failure, because you can't see it. Background tasks can stop running, for example after a migration or an outage, and the agent won't say a word. Data simply isn't collected during that time. | What to do | How | | - | - | | Check recurring results | Did the Friday summary arrive? If not, go to the [Log](/en/agents/maintenance) | | Ask for an explicit sign of life in the task | "If there's nothing to collect, still write one line: checked, no changes" | | Check after every restart | Changing settings and installing skills restart the container | ## It breaks things When an agent acts on its own, it can break something that worked and not realize it: upload data to the wrong place, mess up a page, overwrite a file. And then honestly report that everything is done. | What to do | How | | - | - | | "Show me" first, then "do it" | Until you trust it, use suggestion mode, not action mode | | Keep a [backup](/en/agents/maintenance) | Before any task that touches live data | | Don't give it live data right away | Let it work on a copy or a test page first | | Separate the risks | The agent that handles money and email isn't the one that experiments | ## Access The two most expensive stories aren't about the agent's mistakes but about what it was given: access to email with delete rights, and a skill from an unknown author that contained an instruction to send keys outside. Both are covered in [Security](/en/agents/security). ## How to live with it An agent is an intern, not a genius. A vague task produces confident nonsense. "Collect July expenses from the receipts in the folder, group them by category, send a table, and name the three biggest expenses" works; "sort out the money" doesn't. Five minutes: did the scheduled summaries arrive, what's in the **Stuck** column on the [board](/en/agents/tasks), has anything odd appeared in the files. That's cheaper than discovering a month of lost data. A proven approach: one agent does the work, the second checks the result against the source data. It costs more, but it pays off for critical things. The agent won't come to you and say "I couldn't do it." It will answer your question if you ask. Anything that should arrive on its own, set up as a [scheduled task](/en/agents/schedule) with an explicit requirement to report. ## What's next Rules for granting permissions and third-party skills Log, diagnostics, backup Files that aren't lost during compression How to make the agent show signs of life # Agent Maintenance Source: https://aitextura.mintlify.app/en/agents/maintenance Log, terminal, and container management: restart, diagnostics, agent panel password, backup, and deletion. An agent is a running container, and sometimes you need to do something with it: restart it, see where it got stuck, or take a copy before a risky task. All of this lives in three settings sections: **Log**, **Terminal**, and **Management**. ## Log **Settings → Log** is the agent's technical diary: what it did and where it got stuck. Log section: a live console with technical lines and an auto-refresh toggle You don't need to read it in full, and it wouldn't help. Come here in two cases: the agent is silent and you want to know why, or support asks you to share the log lines. The **Auto** toggle keeps the console live, and there's a manual refresh button next to it. When you contact [support](/en/guides/support), attach the last log lines from the time things broke. It speeds up troubleshooting many times over. ## Management **Settings → Management** covers everything you do with the container as a whole. Management section: the Status block showing Running and the Agent Panel block with a password | Block | What it does | | - | - | | **Status** | Container status plus **Restart** and **Stop** buttons | | **Agent Panel** | Password for the agent's own panel. You don't need it in the workspace, only if you open the panel in another browser | | **Diagnostics** | Checks the container, the model key, channels, and stuck locks | | **Backup** | An archive of the agent's full state: settings, personality, memory, conversations | | **Danger Zone** | Deletes the agent | ### Diagnostics This is the first thing to click when the agent behaves strangely. **Check** only inspects; **Check and Fix** repairs what it finds, and the agent may restart in the process. ### Backup The **Download Backup** button gives you the archive. Make one before any major change: deleting skills, rewriting the personality, or giving the agent access to live data. Diagnostics, Backup, and Danger Zone blocks with a press-and-hold delete button ### Deletion The Danger Zone button works on press-and-hold, not a click, so you can't trigger it by accident. The agent and all its data are deleted permanently: files, memory, conversations. Deleting an agent is the only way to free up a slot if your account already has two deployed agents. Download a backup first: the platform can't restore a deleted agent. ## Terminal **Settings → Terminal** is a command line inside the agent's container. Terminal section: Connected status and a real shell prompt This section is for people who know what they're doing: you can break settings here. If you do, run **Diagnostics → Check and Fix**. You don't need the terminal for everyday work: the agent will do anything you need in the container if you ask in chat. ## If the agent breaks Look for the green dot next to the agent's name at the top of the workspace. No dot means the container isn't running. **Management → Restart**. It takes half a minute and clears most hang-ups. **Check and Fix** finds problems with the model key, a channel, or a stuck lock on its own. If that didn't help, the last log lines show where it's stumbling. Include the log lines and a description of what you did before it broke: [Support](/en/guides/support). ## FAQ None of the files or memory: they're on the container's disk. Only current conversations, scheduled tasks, and connected channels are interrupted, for about a minute. A stopped agent does nothing: no scheduled tasks, no replies. Spending comes from model work, and there's no work. To open the agent panel outside the workspace, for example in another browser. You don't need it inside the workspace. The buttons next to the field show and copy the password. There's no backup upload button in the interface: you keep the archive, and restoring is handled with [support](/en/guides/support). That makes a backup insurance, not a full time machine. Check the log to see whether the task ran at all. After a migration or long downtime, background tasks may fail to start, and the agent won't tell you. See [Where the Agent Fails](/en/agents/limits). ## What's next Common failures and how to catch them Minimal permissions and a backup before risk Fallback models keep the agent from going silent When you've tried everything # Agent Models Source: https://aitextura.mintlify.app/en/agents/models What your agent thinks with: four model slots, a catalog with prices, and connecting your own provider key directly. Out of the box, your agent already thinks and works, so you don't need to touch the models. Come here in two cases: you have your own provider key and want to pay the provider directly, or the agent has started making mistakes and you're looking for a smarter model. ## Four slots **Settings → Models**. Each role has its own model. This isn't duplication, it's a division of labor. Models section: Brain, Fallbacks, Image Reading, and History Compression cards | Slot | What it does | What to consider | | - | - | - | | **Brain** | Thinks and replies. The agent's main model | Defaults to `aitextura-chat`. Change it if the agent can't handle your tasks | | **Fallbacks** | Take over if the main model doesn't respond or refuses | Without fallbacks, a provider outage means a silent agent | | **Image Reading** | Looks at photos you send | The list only shows vision models: there's no point showing a picture to a blind one | | **History Compression** | Condenses a long conversation so it fits in memory | A simpler model works here: it needs to be careful, not smart | The **Fallbacks** slot is the most underrated setting. A provider goes down for ten minutes, and all you see is a silent agent, with no idea why your scheduled tasks failed. ## Catalog The **Select** button opens the list: about a hundred models with search and sorting. Model catalog: a table with Model, Context, and Price columns | Column | What it means | | - | - | | **Model** | The name. The current model has a check mark | | **Context** | How much text the model holds at once. More context means a longer conversation before compression | | **Price** | Input → output per 1M tokens. The first number is the cost to read, the second is the cost to reply | Model card in the catalog: context window, input and output cost, and model ID with a copy button Hover over a model to open a card with details and the **Model ID** field. You'll need it if you connect the same model with your own key. ## Your own model **Settings → General** → **Your Own Model (Your Own Key)** block. The agent will call the provider directly, bypassing the platform's infrastructure, and you'll pay the provider. General section: the Name field and the Your Own Model (Your Own Key) block with provider, key, and model The **Provider** list: who you buy access from. The **Provider Key** field. The key is verified with the provider, isn't stored in the conversation, and is never sent back to the browser. Once you enter the key, the model list loads automatically. You can also type the name manually. The platform checks whether the model understands photos when you connect it, so you don't need to specify that. The platform contacts the provider and confirms the key works. This is also where, in **General**, you change the agent's name: the one shown in the list and in the workspace. ## What it costs The agent spends budget on every action: a conversation, a scheduled task, reading a file, overnight memory cleanup. Keep this in mind when you set up the [schedule](/en/agents/schedule): monitoring every 30 minutes costs dozens of times more than a daily summary. You can see spending and the remaining balance in **Billing** and [Charge History](/en/guides/finance). They're shared across the whole account. ## FAQ First, check the documents on the [Personality](/en/agents/personality) tab and how the task is worded: a vague task produces confident nonsense on any model. Change the model when the task is clearly described and the result still misses. This looks like an outage at the main model's provider. Fill in the **Fallbacks** slot so the agent switches over on its own. You can see what happened in the [Log](/en/agents/maintenance). Check the **Image Reading** slot: if it holds a model without vision, the picture is simply never shown to it. That's history compression. Put a model with a larger context in the **Brain** slot, and keep important facts in memory files: they aren't compressed. It depends on the provider and volume. The practical value of your own key isn't only the price: it's your own limit and your own quota, independent of the account's shared budget. ## What's next What the agent is allowed to do Log, diagnostics, backup The main source of ongoing spending Where the account budget goes # Personality and Memory Source: https://aitextura.mintlify.app/en/agents/personality Your agent's character and everything it remembers about you are plain text files. What each one holds, what's worth adding by hand, and how nightly memory processing works. Your agent behaves the way its files say — and you can open and rewrite those files. There's no black box: the **Personality** tab shows a list of documents on the left and a plain-text editor on the right. The Personality tab: Character and Memory sections on the left, document editor on the right Below the editor are the **Save** button and a note: changes take effect immediately, the agent reads them in new conversations, and no restart is needed. ## Character Five documents describe who the agent is and how it works. | Document | What it controls | What goes in it | | - | - | - | | **Soul** | Character: how it talks, what matters to it, and what it won't do. Reread at the start of every new conversation | Filled in from the role when the agent is created. Edit it when the agent's actual job changes | | **Personality** | Profile card: name, manner, mood | Filled in during the [first conversation](/en/agents/first-run), when you pick the agent's name | | **Rules** | Working rules: what it can do on its own and what requires asking you first | The key document for security. This is where restrictions live | | **About You** | What the agent knows about you: how to address you, language, time zone, what you do | Keep adding to it — the agent will work more accurately | | **Tool Notes** | The agent's own notes on what's connected and how to use it in your setup | Usually maintained by the agent; it helps to add non-standard paths and addresses | The Soul document: character text generated by the platform for the Automator role This is the **Soul** the platform wrote for the "Automator" role: who it is, what it does on first run, which files it maintains. This is the same role you chose in the setup wizard — plain text you're free to rewrite. ## Memory Four documents where the agent's work accumulates. The first two are long-term; the agent maintains the rest as it goes. | Document | What's inside | | - | - | | **Long-Term Memory** | What the agent has committed to memory. You can add to it yourself — the agent reads important facts about your business the next time it checks its memory | | **Dream Journal** | What the agent reflected on overnight and moved into long-term memory while you were away | | `ONBOARDING.md` | The first-meeting script. The agent follows it step by step and then deletes it — if the file is gone, onboarding is complete | | `TASKS.md` | Automation catalog: what it does → where the script lives → how to run it → schedule | | `WORKING.md` | Work log: in progress, plan, done. The only file the agent rewrites in full; it appends to all the others | The TASKS.md document: a prefilled Automation Catalog template with Running and Idea Queue sections `TASKS.md` is the most useful file for you, not the agent. It lists everything you've automated and how each piece runs: open it a month from now and you'll understand your own setup without asking the agent. ## Nightly processing A separate item under the Memory section opens the **Background Self-Improvement** panel. While you're away, the agent rereads the day's conversations, pulls out what matters, and moves it into long-term memory — from the **Dream Journal** into **Long-Term Memory**. The Background Self-Improvement panel for a newly created agent — empty For a new agent, the panel is empty: no nights have passed yet. Check it when the agent starts forgetting agreements — you'll see what it considered important and what it lost. ## What to add yourself The agent writes its own files, but three things are safer to enter by hand — they shouldn't depend on what the agent decided was important. | Document | What to add | | - | - | | **Rules** | What it must not do without asking: message people, publish, delete files, spend money | | **About You** | Your time zone, how to address you, what your business does, who your customers are | | **Long-Term Memory** | Permanent facts: company details, project names, colleagues' names, links to working spreadsheets | Anything said only in chat lasts until the history is compressed — long conversations get condensed and details disappear. Any agreement you can't afford to lose belongs in a file. ## FAQ No. Changes are picked up in new conversations. If an edit didn't take effect, start a new conversation: the current one still holds the old version. Working memory is finite, and long conversations get compressed — details are lost in the process. Add the fact to **Long-Term Memory** by hand. Also check that the **Memory** ability is enabled under [Abilities](/en/agents/skills). Check **Long-Term Memory** and the **Dream Journal**: a wrong conclusion may have landed there during nightly processing and is now treated as true. Delete the incorrect line — you fix this by editing the file, not by arguing in chat. The **Personality** document holds the name, manner, and mood. The **Soul** document holds what matters to the agent and what it won't do. Change tone in the first, behavior in the second. Yes, the agent understands both. Avoid mixing languages within one document — it muddles the workflow itself, not the agent's understanding of the words. No. Some documents are empty for a new agent: **Personality**, **Rules**, and **About You** are filled in during the first conversation or by you. An empty document shows the hint "Describe what the agent should be like — in plain text or Markdown." ## Advanced You don't need this to get started. Documents support Markdown: `##` headings, lists, bold text. This helps the agent — it uses headings to find the right section without rereading the whole file. Keep one topic per document: rules in one place, facts about you in another. It's a log of current work, not an archive: "in progress," "plan," "done." If the agent only appended to it, the file would keep growing and eat up memory. The agent appends to all the other documents. Yes, the [Files](/en/agents/files) tab shows them as regular container files alongside everything else. The **Personality** tab is a showcase for the documents you edit most often. You can control what condenses long conversations: the [Models](/en/agents/models) section has a dedicated **History Compression** slot. This model needs precision more than intelligence — but one that's too weak loses more details. ## What's next What the agent is allowed to do and which skills it has installed The agent's entire working environment Separate personas with their own character inside the agent What the agent thinks with and what compresses its history # Create an agent Source: https://aitextura.mintlify.app/en/agents/quickstart A three-step wizard: pick a role, pick an engine, wait for deployment. The agent is ready to work in about a minute. You don't need to set up a server or configure an environment: you pick a role from a ready-made list, and the platform deploys a container and writes the agent's personality for you. From clicking the button to the first message takes about a minute. ## Hire an agent The **Agents** section in the left menu is a shared list of all agents and AI Employees on the account. The Agents section: a table with Name, Status, Created, Channels, Talents, and Last Activity columns, and the Hire Agent button | Column | What it shows | | - | - | | **Name** | The agent's name; click it to open its workspace | | **Status** | Running · Active · Paused | | **Channels** | Icons of connected messengers | | **Talents** | How many personas and skills are connected | | **Last Activity** | When the agent last did something | | **Live chat** | Opens the conversation right from the list | On the right of each row is an on/off toggle and a `...` menu with actions, including delete. At the top are a name search and the blue **+ Hire Agent** button — that's where it all starts. The first wizard step, **"What are we creating?"**, has two cards: **AI Employee** (a bot for customers) and **Autonomous Agent** (an independent worker with its own workspace). Click the second one. The “What are we creating?” step: AI Employee and Autonomous Agent cards The **"Where do we start?"** step shows ready-made roles, each with a predefined personality and workflow. Six cards are visible; the **Show more** link expands all 25. The “Where do we start?” step: the first six role cards The **From scratch** link below the list creates an agent without a preset — you'll have to write its name and personality yourself. For the full list of roles and what they do, see [Roles](/en/agents/roles). The **"Set up the agent"** step. The **Name** field is prefilled with the role name; you can change it. Below it, choose the engine. The “Set up the agent” step: OpenClaw or Hermes engine choice and the Name field | Engine | What it means | | - | - | | **OpenClaw** | The agent has its own web interface — you open it via a link and see the full workspace. Selected by default | | **Hermes** | No web interface: all communication and management go through messengers | Click **Create agent →**. Right after creation, the platform offers to connect Telegram — no token needed; the bot is created in a couple of taps. You can put this off: the **Skip for now** link moves on, and you can connect a channel later from the workspace. Details: [Connect](/en/agents/connect). The **"Deploying the agent"** screen shows a timer and five stages: starting the container → preparing the workspace → connecting tools → setting up work rules → final checks. The “Deploying the agent” screen: five stages, some with green check marks In our run, this took 55 seconds. You can close the page — the platform will email you when the agent is ready. The “Automator is ready to work” screen with a green check mark and an offer to connect Telegram ## Check it Open the agent and click **Chat** in the top-right corner of the workspace — a side panel with the conversation opens. The chat panel: “Automator is online,” an input field, and the “What can you do?” hint Type "Hi". A fresh agent won't reply with a canned phrase — it starts getting to know you: asks how to address you and suggests a name for itself. This is the [first run](/en/agents/first-run) — it's written in a file on the agent's side and runs once. A green dot next to the agent's name at the top means the container is running. If the dot isn't green, check [Maintenance](/en/agents/maintenance) — it has restart and diagnostics. ## Agent limit An account can have **no more than two agents** deployed at a time. If you try to create a third, the platform shows a warning and won't let you continue. To free up a slot: **Agents** → the `...` menu of the agent you don't need → **Delete** → confirm by typing its name. After that, the creation wizard works again. Deletion is permanent: the agent's files, memory, and conversations go with it. If you still need the data, download a [backup](/en/agents/maintenance) first. ## FAQ Keep **OpenClaw** — it's selected by default and gives the agent its own web interface that opens via a link. Hermes is for when you don't need a web interface at all and all communication goes through a messenger. No. A role is starting personality text, not a hard setting: open [Personality](/en/agents/personality) and rewrite the **Soul** document for your task. Changes take effect immediately; no restart needed. Wait for the email — you can close the page; the process runs on the platform side. If the agent never shows up in the list, contact [Support](/en/guides/support). Yes, click **Skip for now**. You can talk to the agent right in the workspace using the **Chat** button. Telegram can be connected later on the [Connect](/en/agents/connect) tab — its main purpose is getting results on your phone. Check the status: a green dot next to the name means the container is running. If the agent was just restarted or you changed settings that require a restart, wait 20–30 seconds. Next step: **Settings → Management → Diagnostics**. ## What's next What the agent asks in the first conversation and how to answer Connect Telegram and manage the agent from your phone Personality and memory files: what's in them and what to add 25 ready-made roles and what each one does # Agent roles Source: https://aitextura.mintlify.app/en/agents/roles 25 ready-made roles in the creation wizard: the name and personality fill in automatically. Roles by area and what to do if none of them fits. A role saves you from a blank page: the platform writes the agent's personality, workflow, and first questions for you. You pick it on the second wizard step, **"Where do we start?"**. Six cards are shown by default; the **Show more** link expands all 25. The expanded list of roles in the agent creation wizard ## Personal and organization | Role | What it does | | - | - | | **Personal Assistant** | Handles your affairs: tasks, agreements, drafts | | **Morning Briefing** | Sends a short brief every morning: the day's plan, deadlines, and loose ends | | **Second Brain** | A knowledge vault: you send links, articles, and thoughts — it files them in the right place | | **Secretary** | Meeting minutes, business emails, tasks, and contacts | | **Meeting Minutes Taker** | Turns meeting notes and transcripts into clear minutes | | **Finance Assistant** | Tracks expenses and income, budgets, and gives honest summaries of where the money went | | **Habit Coach** | Morning checklist, evening review, and tracking of what's sticking | ## Content and social media | Role | What it does | | - | - | | **Copywriter** | Writes posts, emails, and landing pages in your voice | | **SMM Manager** | Runs the content plan: ideas, posts, and captions for your platforms | | **Content Factory** | A publishing pipeline: keeps the calendar and puts out content in a steady stream | | **TikTok Factory** | Short videos: ideas, hooks, and second-by-second scripts | ## Customers and sales | Role | What it does | | - | - | | **Sales Manager** | Customer emails, scripts, objection handling, and follow-ups | | **Customer Onboarding** | Guides each new customer through the launch stages: welcomes, checklists, reminders | | **Review Replies** | Drafts replies to reviews — from thank-yous to handling complaints | | **Customer Support** | Answers from your materials and hands complex cases to a human | | **Meeting Dossier** | Prepares you for negotiations: compiles a dossier on the person and the company | The **Customer Support** role answers you, not your customers directly. For live conversations with customers in channels, you need an [AI Employee](/en/guides/ai-employees) — it has a Knowledge Base, Self-Learning, and Working Hours. ## Data and monitoring | Role | What it does | | - | - | | **Researcher** | Gathers information and turns it into short summaries | | **Report Analyst** | You send exports and numbers — you get a clear report | | **Trend Radar** | Scans sources in your niche every morning and sends a digest | | **Mention Radar** | Tracks what people say about your brand and competitors | | **Price Radar** | Tracks competitor prices on specific pages and keeps a history | ## Documents and learning | Role | What it does | | - | - | | **Automator** | Turns routine work into scripts in its own workspace | | **Contract Review** | Reads contracts for you and translates legalese into plain language | | **Recruiter** | Writes job postings and compares résumés against requirements | | **Tutor** | Explains things in simple terms, keeps a study plan, and tracks progress | ## How to choose Pick a role by the action, not the job title. Need a regular report — **Report Analyst**. Need file processing and scripts — **Automator**. Need one short text every morning — **Morning Briefing**. A role doesn't restrict anything: it sets the starting personality text and the first conversation. An agent with the Tutor role won't refuse to tally your expenses — it'll just do it in a tutor's manner, and worse than an agent set up for that. The **From scratch** link below the list creates an agent with no personality. This only makes sense if you're ready to write the **Soul** document yourself — otherwise the agent will ask follow-up questions at every step. You can always edit the written personality: [Personality](/en/agents/personality). ## FAQ There's no separate "change role" button — the role becomes text on the **Personality** tab. Open the **Soul** document and rewrite it for the new task: that's how you change the role. While you're at it, clean up **Long-Term Memory** if the agent has already picked up facts from its previous work. Technically yes, practically it works poorly. Working memory is finite: an agent with two areas of responsibility starts mixing up contexts and ignoring its own rules. It's better to deploy a second agent (you're allowed two) or add a [persona](/en/agents/team) to the current one. Take the closest one by type of action and rewrite the **Soul** in your own words. That's faster and more reliable than creating an agent **From scratch**: a ready-made role already has working habits built in — keeping a log, maintaining a task list, asking questions one at a time. ## What's next Three wizard steps and deployment in a minute How a role becomes text you can rewrite Five areas with example task wording Several personas with different personalities inside one agent # Schedule Source: https://aitextura.mintlify.app/en/agents/schedule The agent's scheduled jobs: a morning summary, a Friday report, an hourly price check. The agent runs them on its own, even when the Agent Workspace is closed. The most valuable thing about an agent is that it works without reminders. Explain once — "every Friday, send me a summary of expenses and leads" — and from then on it does this on its own, even when the Agent Workspace is closed and you're asleep. ## Create a job Open the **Schedule** tab. For a new agent, the list is empty and the creation form is right below it. The Schedule tab: an empty job list and the New job form | Field | What to enter | | - | - | | **Name** | A short job name: "Morning summary" | | **When to run** | A frequency from the list, plus a time | | **What to do** | The task in words — this is the exact text the agent receives when the job runs | | **Send result to channel** | On by default: the result arrives in your connected messenger | The **What to do** field matters most. Write it the way you'd write a task in chat: "Gather the latest news on our topic and send a short summary — five points, one line each." ## Frequency The When to run dropdown open with all options | Option | When it's useful | | - | - | | **Every day** | Morning brief, evening review | | **Weekdays (Mon–Fri)** | Work summaries that shouldn't arrive on weekends | | **Once a week** | Weekly report, plan for next week | | **Every hour** | Monitoring: prices, mentions, new leads | | **Every 30 minutes** | Same, when response speed matters | | **Custom schedule** | Everything else | Times follow the time zone the agent has on record for you. Tell the agent your time zone in conversation and also add it to the **About You** document on the [Personality](/en/agents/personality) tab — otherwise your 09:00 summary may arrive on a different morning than you meant. ## Your first three jobs A common starting point. | Name | When | What to do | | - | - | - | | Morning brief | Every day, 08:00 | Weather, calendar meetings, important unread emails, and anything urgent on the task list — in one message | | Weekly wrap-up | Once a week, Friday 18:00 | Expenses by category, new leads, what's still open — plus a plan for next week | | Monitoring | Every hour | Check competitors' pages and message me only if something changed | Add the condition "message me only if there's something to report." Otherwise hourly monitoring turns into twenty-four "no changes" messages a day, and you stop reading them. The full New job form: the Send result to channel toggle and the Create button ## Verify Scheduled jobs are where the agent can go quiet without you noticing: a job stops running, and since you don't expect messages every day anyway, you miss it. * After creating a job, set it to run soon and watch the first run happen live. * Once a week, check that everything that should arrive is arriving. * If it's silent, open the [Log](/en/agents/maintenance) — it shows whether the job ran and where it failed. ## FAQ This usually means the container is restarting — for example, you just enabled the task board or toggled a skill. Wait 20–30 seconds and click **Retry**. It won't: the agent stays silent about failed background jobs. Check the [Log](/en/agents/maintenance) and, if the container restarted, run **Diagnostics**. For critical jobs, make a habit of checking results against the calendar. Check the **Send result to channel** toggle in the job, and make sure the channel itself is connected on the [Connect](/en/agents/connect) tab. The agent calculates time using your time zone. If it doesn't know it or has it wrong, fix the **About You** document on the [Personality](/en/agents/personality) tab. The interface doesn't limit the number of jobs, but every run spends your model budget. Monitoring every 30 minutes costs noticeably more than a daily summary — keep that in mind when you set things up. ## What's next One-off tasks on the board Where job results are delivered Log, diagnostics, and restart What the agent thinks with and what it costs # Security and Access Source: https://aitextura.mintlify.app/en/agents/security Five rules for working with an autonomous agent: minimal permissions, separate accounts, no skills from strangers, and a "show me first, then do it" mode. Your agent has a command line, access to your files, and access to whatever services you've opened to it. That's exactly what makes it useful, and exactly what makes mistakes expensive. The five rules below prevent most of the pain. ## Five rules | Rule | Why | | - | - | | **Separate accounts, not your personal ones** | Give the agent its own email, its own spreadsheet access, its own account in each service. Then any mistake stays within that account | | **One task, one agent** | Different areas of responsibility mean different access and different risk. Don't put experiments next to money | | **Skills only from well-known authors** | A third-party skill is text with instructions that the agent will follow without a second thought | | **"Show me" first, then "do it"** | Permissions grow as trust grows, not in advance | | **The agent is an intern, not a genius** | It needs clear procedures and verifiable results, or you'll get confident nonsense | The rule "don't install the agent on your personal computer," familiar from self-hosted setups, doesn't apply here: the agent lives in a container on the platform side. Nothing runs on your machine, and the agent can't see your files, only what you've given it access to yourself. ## What not to give | Don't give | Give this instead | | - | - | | Access to your personal email with delete rights | A separate mailbox or read-only access | | Keys to wallets, banking, or payment services | Nothing. Don't assign financial operations to an agent | | Admin passwords for production systems | An account with permissions scoped exactly to the task | | Access to the customer database "just in case" | An export of the part you need, in a file | Anything the agent can do, it will eventually do by mistake, including irreversibly. Base your permissions on that, not on how well it has behaved so far. ## Who can message the agent The only barrier between the agent and outsiders is the list of Telegram IDs on the [Connect](/en/agents/connect) tab. Anyone on the list controls the agent: its files, its command line, its access. * Only add people you trust with that access in full. * When an employee leaves, remove their ID from the list. * Shared access "for the team" is a reason to create a separate agent with separate permissions. If anyone besides you messages the agent, turn off the **Command Execution** ability in [Abilities](/en/agents/skills). That's the first thing to lock down. ## Skills from the catalog A skill is a plain text file with instructions. It can hide a command to find keys and send them to a third party, and the agent will follow it without a second thought: to the agent, it's just part of the instructions. A well-known author, thousands of installs, a recent update. This lowers the risk but doesn't remove it. The skill is on the [Files](/en/agents/files) tab and opens as text. Anything that sends data outside is a reason to pass. You can generate the skill you need for your task. That's more reliable than installing someone else's file for a single feature. Sometimes the agent refuses to install a suspicious skill on its own and tells you so. Treat that as luck, not a safety system: the decision is still yours. ## What to put in Rules The agent rereads the **Rules** document on the [Personality](/en/agents/personality) tab in every new conversation, unlike what you say in chat, which is lost when history is compressed. A minimal set: * Don't message people or send emails without my confirmation. * Don't publish anything externally without my confirmation. * Don't delete files or emails; move them to an archive. * Don't spend money or place orders. * Only put things in the `site/` folder that can be shown to outsiders. * Before a task that changes live data, show the plan and wait for my reply. ## Before a risky task **Settings → Management → Backup**. An archive of the agent's full state: [Maintenance](/en/agents/maintenance). "Describe step by step what you're going to do, and don't start yet." Half the problems show up right here. One folder, ten records, a test page. Once you've accepted the result, expand. ## FAQ No. It runs in its own container on the platform side. It can only access what you put there or what you've given it access to. Conversations and files live in the agent's container. It can be controlled by anyone with access to your AI TEXTURA account and by the people whose Telegram IDs you added on the **Connect** tab. Only from a backup, if you have one. That's exactly why you should add a no-delete rule to **Rules**: have it move things to an archive instead of deleting them. Revoke keys and tokens in the services themselves, change passwords, remove extra Telegram IDs on the **Connect** tab, and if needed, stop the agent in **Management**. Then contact [Support](/en/guides/support). No. That's the job of an [AI Employee](/en/guides/ai-employees): it has a Knowledge Base, response quality control, and working hours. The agent has none of that, but it does have a command line, so don't let outsiders near it. ## What's next Four failures no setting can fix What to turn off if others have access The list of people who can message the agent Backup, diagnostics, stopping the agent # Agent Sites Source: https://aitextura.mintlify.app/en/agents/sites Everything the agent puts in its site/ folder goes live on the internet immediately — no deployment and no server of your own. Main site, additional sites, and a custom domain. Ask in chat for "a page with a lead form" and get a working link a minute later — this is where that happens. Everything in the agent's `site/` folder is available online as soon as the file is saved: no deployment, no server, no hosting setup. The Sites tab: the main site with a Populated status and a public link, plus the Additional sites and Custom domains sections ## Main site The root of the `site/` folder opens at the agent's main address, which looks like `site-.agents.aitextura.com`. The **Open** button takes you to the live page, and the badge next to it shows whether the site has content or is empty. The easiest way to change it is in words: "remove the second section and move the booking button to the top." The agent edits the files itself, and the update is visible right away. ## Additional sites Each subfolder inside `site/` is a separate site with its own address. The **+ Create site** button creates a folder with a placeholder page and gives you a live address right away. This keeps different things apart: a landing page, an internal metrics dashboard, a page for a contractor. ## Custom domain The **Custom domains** section connects your own domain to any of the agent's sites. One CNAME record pointing to the agent's address. Nothing else to configure. In the **Domain** field, enter something like `www.example.com`. In the **Which site** list, choose which of the agent's sites to serve at that address. The certificate is issued automatically on the first visit — no need to set up HTTPS by hand. ## What's public The entire `site/` folder is public. Anything the agent puts there is available to anyone who knows the address: drafts, exports, customers' personal data. Working files belong in `workspace`, not in `site/`. State this explicitly for the agent in the **Rules** document on the [Personality](/en/agents/personality) tab: "only put things in site/ that outsiders are allowed to see." ## FAQ Reload the page with the browser cache cleared. If that doesn't help, check on the [Files](/en/agents/files) tab that the agent actually saved the file in `site/` and not next to it. The interface has no separate access setting. If a page isn't meant for outsiders, don't keep it in `site/`: share the file in chat or put it in `workspace`. Check the CNAME record at your registrar and wait: DNS updates take anywhere from a few minutes to a few hours. The certificate is issued on the first successful visit; until then, the browser may show a warning. Ask it to restore the previous version — it keeps its changes in files. It's safer to have a [backup](/en/agents/maintenance) and not let the agent edit a live page without your review: first "show me what you're going to do," then the action. ## What's next Where site/ lives and what's around it Page edits as cards on the board Other use cases where the agent writes code What must never be made public # Abilities and Skills Source: https://aitextura.mintlify.app/en/agents/skills What your agent is allowed to do: core abilities with toggles, installed skills, and the catalog. Plus why a third-party skill is a risk. Your agent can read files, run programs in its container, and go online, and each of these can be turned off with a single toggle. **Abilities** set the limits of what it can do at all: if an ability is off, the agent won't use it, no matter how nicely you ask. ## Core abilities **Settings → Abilities**. The top three abilities can be toggled; the rest are always on. Abilities section: Files, Command Execution, and Memory toggles, plus always-on Internet, Interface, and Conversations | Ability | What it gives | Turn it off when | | - | - | - | | **Files** | Read and create files in its own storage | Almost never: without files, the agent loses its memory of the work | | **Command Execution** | Run programs inside its container | This is the first thing to turn off if anyone else is messaging the agent | | **Memory** | Remember important things between conversations | Only if you need a one-off agent that doesn't build up knowledge about you | | **Internet** | Search for information and open websites | Always on | | **Interface** | Open websites in a browser and show images | Always on | | **Conversations** | Search past conversations | Always on | **Command Execution** is the most powerful and most dangerous ability: it's a full command line. If anyone besides you has access to the agent, turn it off and enable it only for specific tasks. ## Installed skills Further down the same tab is the list of skills. A new agent comes with several dozen. A skill is an instruction file: the agent reads it and learns something new. List of installed skills with Ready status and toggles | What you see on the card | What it means | | - | - | | Skill name | The technical name, usually in English | | **Ready** | The skill works and is ready to use | | **Setup Required** | The agent is missing something, most often access or a key | | Toggle | Turns the skill on and off | | Trash can | Deletes the skill | Toggling a skill restarts the agent: chats and scheduled tasks pause for about a minute. The **Filter installed skills** field searches the list. With several dozen skills, it's faster than scrolling. ## Catalog **Settings → Catalog** has new skills and plugins. There's a **Skills / Plugins** switch, a category filter, and **Top** and **Trending** collections. Skills catalog: cards with author, rating, install count, and an Install button Each card shows the author, rating, install count, and last update date. The **Install** button installs the skill, and the agent picks it up after a restart. A skill is text with instructions that the agent follows without a second thought. A downloaded file can hide a command to collect your keys and send them to a third party. Don't install skills from unknown authors: it's safer to ask an AI to write the skill you need from scratch for your task. Learn more in [Security and Access](/en/agents/security). What to check before installing: a well-known author, lots of installs, a recent update. This lowers the risk but doesn't remove it, so read what's inside. ## FAQ Tell it directly: "use skill such-and-such." The agent picks its tools on its own and doesn't always guess right. If it says it's missing data, check the status: **Setup Required** means some access is missing. It's restarting, so wait half a minute. If it doesn't come back, turn off the new skill and run **Diagnostics** in [Maintenance](/en/agents/maintenance). There's no full guarantee. Look at the author, install count, and update date. You can also open the file itself on the [Files](/en/agents/files) tab and read it: it's plain text. Anything that asks to send data outside is a reason to pass. That happens: it may find a skill suspicious and tell you so. That's a good sign, but don't rely on this check. The decision is still yours. They're similar in purpose: both extend what can be done. But [talents](/en/guides/talents) connect an AI Employee to your external systems and are configured at the AI Employee level, while skills live inside a specific agent's container. ## What's next Five rules that prevent most of the pain What the agent thinks with Where skills physically live What to do if a skill broke everything # Agent Tasks Source: https://aitextura.mintlify.app/en/agents/tasks The agent's kanban board: you add a task, the agent picks it up. How to enable the board, what the eight columns mean, and how to create cards. You can assign tasks in chat, but then they exist only in the conversation: it's hard to tell what the agent picked up, what's stuck, and what's done. The **task board** is a shared to-do list where each task is a card and you can see which stage it's in. ## Enable the board The board is off by default — the **Tasks** tab shows an **Enable board** button. The Tasks tab with the board disabled and an Enable board button The platform warns you that the agent needs to restart: chats, automations, and connected channels will be interrupted for about a minute. The Agent will restart dialog with Cancel and Enable board buttons Right after you enable it, the board may show "Couldn't load the task list" — the container is still restarting. Wait 15–20 seconds and click **Retry**. ## Columns Eight columns, left to right, trace a task's path from idea to result. The task board: Ready, In progress, In review, Stuck, and Done columns | Column | What's in it | | - | - | | **Triage** | The task is recorded but not yet defined as work | | **Queued** | Triaged and waiting its turn | | **Waiting** | Blocked: missing access, data, or your reply | | **Ready** | Can be picked up | | **In progress** | The agent is working on it right now | | **In review** | There's a result, and it needs your eyes | | **Stuck** | The agent couldn't finish — open the card to see why | | **Done** | Closed | The **Dispatch** button in the top right moves queued tasks into work in bulk — this is your "go ahead" command. The **Columns / List** toggle switches the board view, and the **Hide empty columns** checkbox removes clutter: with only two or three tasks, eight columns just get in the way. ## New task Click **+ Card** in the top right, or **+** in the header of the column you want. The New task dialog: What needs to be done, Details, Priority, Labels, and Column fields | Field | What to enter | | - | - | | **What needs to be done** | The task title in one line, as in a to-do list | | **Details** | Optional, but this is where the agent learns what counts as a finished result | | **Priority** | Defaults to "Normal" | | **Labels** | Comma-separated: `website, urgent`. You can filter the board by them later | | **Column** | Where to put the card. If you don't pick one, it goes to the default column | Phrase the task as an outcome, not a direction: "put July expenses into a spreadsheet by category and send me the file" instead of "sort out the expenses." ## How it works The board doesn't run the agent by itself — it gives the agent a queue to pull work from. From the Agent Workspace or right from chat: ask the agent to create a task, and it will make the card itself. Click **Dispatch** or move the card to **Ready**. The card moves to **In progress**, and the result goes to **In review**. If you accept it, close the card. If the agent couldn't finish, the card ends up in **Stuck** with the reason. ## FAQ Queued tasks don't move into work on their own — click **Dispatch** or drag the card to **Ready**. If nothing happens after that, check the container status: the green dot next to the agent's name. Open the card in the **Stuck** column: the agent writes down where it stopped. Most often it's missing access, or the description leaves out the key thing — what counts as a finished result. This happens: the agent may report success after completing only part of the work. Verify the actual result — the file, the email, the row in the spreadsheet. Learn more in [Where the agent breaks](/en/agents/limits). Wait half a minute and click **Retry**: the task list is unavailable while the container is starting up. If that doesn't help, go to **Settings → Management → Diagnostics**. The board holds one-off tasks you add by hand. The [Schedule](/en/agents/schedule) holds recurring jobs the agent runs on its own at set times. ## What's next Jobs the agent runs on its own at set times Assign tasks and get reports from your phone Different personas for different kinds of tasks Why you should verify the actual result # Agent Team Source: https://aitextura.mintlify.app/en/agents/team Personas are additional workers inside a single agent: each with its own character, its own conversations, and, optionally, its own Telegram bot. An agent juggling five areas of responsibility performs worse than one focused on a single area — working memory is finite. **Team** solves this inside the same agent: you create a persona for each job, and each one has its own character and its own conversations. The Team tab: the Automator persona card with a Main badge and the Add persona button An agent persona is not the same as an [AI Employee's Persona](/en/guides/persona). There, it's the instructions the AI Employee follows when talking to customers; here, it's a separate worker inside the agent. ## Persona card A new agent has one persona — the main one, marked with a **Main** badge. Each card has three buttons. | Button | What it opens | | - | - | | **Workspace** | This persona's files | | **Chat** | A conversation with this persona specifically | | **Personality** | Its character and memory — the same documents as on the [Personality](/en/agents/personality) tab | The **+ Add persona** button creates a new one: give it a name, describe its character, and from then on it works as a separate worker, but inside the same container. ## Why split the work The classic setup is a pipeline where everyone has their own job. | Persona | What it does | | - | - | | Researcher | Gathers news and materials on a topic | | Writer | Turns them into text in your style | | Reminder | Sends the finished result to Telegram at the right time | You can give each one its own Telegram bot — then in the messenger they appear as different contacts, and you message whichever one you need directly. ## Persona or a second agent | | Persona | Second agent | | - | - | - | | Working environment | Shared with the other personas | Its own, fully separate | | Files and skills | Shared across the container | Its own | | Limit | The interface doesn't limit the number of personas | No more than two deployed agents per account | | When to choose | Tasks in one area that work with shared materials | Different areas, different access, different risk | The rule is simple: different roles in the same job — personas. Different jobs — different agents. Personas share one container: they have the same files, the same skills, and the same access. Don't put a persona that handles email and money next to one you use for rough experiments. ## FAQ Each has its own conversations, but files and the working environment are shared. To hand work between personas, use a file in `workspace`. Yes, each can have its own Telegram bot. You create it the same way as the main one — on the [Connect](/en/agents/connect) tab. The main persona is the agent itself, so it always stays. The only thing you can change is its character: the documents on the [Personality](/en/agents/personality) tab. Exactly as many as you have distinct roles in one job. A persona you talk to once a month only adds confusion — it's easier to keep that task with the main one. ## What's next A persona's character is made of the same text documents A separate bot for each persona A shared board that shows all the work What the container the personas live in can do # What to assign an agent Source: https://aitextura.mintlify.app/en/agents/use-cases Five areas where people usually start: content, money, email and requests, scheduled work, and code. With example task wording. The hardest part of having an agent is figuring out what to hand it. Below are five areas where it pays off fastest, with real examples of task wording you can send to the chat almost word for word. The general rule: hand over what you do manually on a regular basis, following the same routine. A one-off task from scratch is work for you, not for the agent. ## Content The agent keeps an idea bank and turns ideas into ready-made posts. | What you do | What the agent does | | - | - | | Record your thoughts as voice messages throughout the day | Transcribes them and adds them to the idea bank | | Write "I need a post about X" | Pulls an idea, expands it into text, and picks images | | Share your channel archive | Learns from it and writes drafts in your style | | Ask it to run a publishing calendar | Keeps the plan and reminds you what goes out tomorrow | > Take my voice messages from this week, pull three post ideas, and expand the best one into a 1,500-character post. ## Money The bookkeeping people usually put off until the end of the month. | What you do | What the agent does | | - | - | | Send photos of receipts and invoices | Sorts them into categories and keeps a spreadsheet | | Ask where the money went | Sends a weekly summary and shows where the budget slipped | | Provide a list of customers and payments | Tracks outstanding balances and reminds those who haven't paid | > Every Friday at 6:00 PM, send me the week's expenses by category and a separate list of unpaid invoices. ## Email and requests Your inbox sorted without you. | What you do | What the agent does | | - | - | | Give access to your email | Filters out newsletters and surfaces what needs a reply today | | Show your standard replies | Drafts responses from your template — all you do is hit send | | Forward a new request | Logs it in a spreadsheet, sets a status, and reminds you if it's been silent for three days | ## Schedule Explain it once, and from then on it does the job on its own, without reminders. This is what sets an agent apart from any AI chat. | Job | When | | - | - | | Morning brief: weather, today's meetings, important emails, urgent tasks | Every day at 8:00 AM | | Summary of expenses, requests, and tasks for the next week | Fridays | | Latest news and mentions on your topic | Every morning | | Price check on competitor pages | Every hour | Set this up on the [Schedule](/en/agents/schedule) tab. ## Code and sites The agent writes code, publishes it, and keeps it running — no server or deployment on your side. * Everything it puts in the `site/` folder is instantly available online at the agent's address — you can ask it to build a landing page or an internal dashboard right in the chat. See [Sites](/en/agents/sites). * Small tools for yourself — a task manager, a calculator, a request form — get built through a regular conversation in Telegram. * Recurring scripts: processing exports, scraping pages, converting spreadsheets. ## What else people assign The agent is connected to work chats and checks managers' replies against your guidelines around the clock. It's a QA department that never gets tired and never goes easy on colleagues. A real example: ten clubs with courts, no free slots. The agent checked the websites every few minutes and grabbed a slot the moment someone canceled, then messaged the coaches and agreed on a time itself. Swap the court for a supplier and the coach for a contractor — the mechanics are the same. Before you put money into development, the agent combs through open repositories and the market to assess how crowded the niche is: a list of existing solutions often saves the team weeks of work. Once you have more than one agent, split them by topic: one collects data, another writes text, a third sends reminders. Within a single agent, [personas](/en/agents/team) do the same. Tracking meals from a photo of your plate, a voice-note journal, a study plan from textbook photos, a schedule for kids' activities. The limit here isn't technical: just don't pile everything onto one agent. ## Where to start Pick one task you do manually every week and hand it over completely. Not five at once — [memory is finite](/en/agents/personality), and an agent with five areas of responsibility performs worse than one with a single area. The ready-made roles in the wizard are already set up for these scenarios: **Automator**, **Finance Assistant**, **SMM Manager**, **Morning Briefing**. See the full list in [Roles](/en/agents/roles). ## What's next Three wizard steps and deployment in a minute 25 ready-made roles — personality and first tasks fill in automatically Scheduled jobs: the agent runs them without you What it does poorly and how to catch it # amoCRM Source: https://aitextura.mintlify.app/en/channels/amocrm Connect amoCRM as a channel: your AI Employee replies to customers right from the deal card, and you choose which sources and pipelines it handles. Customers write in WhatsApp, Telegram, Avito, and VK, and your manager sorts it all out in amoCRM. The **amoCRM** channel routes those same messages to your AI Employee: it replies right in the deal card, the conversation stays where it is, and nobody has to learn a new interface. ## Connect the channel Open the AI Employee → **Channel** tab → **Connect New** → **amoCRM**. The Add Channel panel with the amoCRM platform selected and the CONNECT AMOCRM button Enter a channel name (only you see it) and click **CONNECT AMOCRM**. An amoCRM window opens: sign in to your account and grant access. The platform handles the rest, and the channel appears on the **Connected** tab. ## Choose sources To open a connected channel, click it on the **Connected** tab. The first thing to do is narrow the scope: your AI Employee doesn't always need to reply in every chat at once. The Edit Channel window: channel name and the expanded Sources list with checkboxes for connected messengers The **Sources** list shows the messengers set up in your amoCRM: Telegram, Avito, VK, Facebook, and others. Uncheck the ones the AI Employee shouldn't handle. For example, leave it Avito and Telegram, and let your managers reply in VK themselves. **Pipeline Filter** does the same for pipelines. While nothing is selected, the AI Employee replies to all deals. Select one pipeline, and it works only there, leaving your cold leads or post-sale support alone. ## Co-Pilot mode The Co-Pilot Mode toggle with a warning that AI replies are posted as notes on the deal With **Co-Pilot** on, the AI Employee sends nothing to the customer. Instead, it writes its reply as a note on the deal: the manager reads the suggestion, edits it as they see fit, and sends it themselves. This is the best way to roll out an AI Employee in a live sales team: for a week it writes notes, you review what it suggests and refine the [Persona](/en/guides/persona), and only then do you let it talk to customers. ## Settings The Pipeline Filter and Settings blocks: message buffer with buffer time and history auto-cleanup **Message Buffer** makes the AI Employee wait before replying, since customers often split one thought into several messages. In a CRM, set a longer buffer than in a messenger, up to 120 seconds: messages arrive here with a delay, and an extra minute of silence doesn't bother anyone. **History Auto-Cleanup** deletes the conversation on a schedule. Auto-cleanup deletes messages permanently. ## Actions in amoCRM The channel only brings in messages; on its own, it doesn't create deals or tasks. To let your AI Employee work with cards hands-on (create deals and contacts, look up customers, add notes and tasks), connect the amoCRM talent on the **Talents** tab: [here's how](/en/guides/connect-talents#how-to-connect-amocrm). ## FAQ The channel is connected under that manager's user account. Create a separate user in amoCRM, such as "AI TEXTURA," and reconnect the channel under it: from then on, replies will go out under that name. Webhooks from a previous connection are still there. In amoCRM, delete the AI TEXTURA integration entirely along with its webhooks, then delete the channel on the platform and connect again. First, check the robots and automation rules in amoCRM itself: they're more often what closes the deal, not the AI Employee. If there are no such rules, contact [Support](/en/guides/support) and include an example deal. ## What's next The same setup for Bitrix24 Deals, tasks, and notes handled by your AI Employee All the ways to connect customers to your AI Employee What the AI Employee remembers about a customer # Bitrix24 Source: https://aitextura.mintlify.app/en/channels/bittrix24 Connect Bitrix24 as a channel: your AI Employee replies to customers in your portal's Open Channels, and you choose pipelines, reply mode, and operator involvement. Customers write in WhatsApp, Telegram, Avito, and your website chat, and your managers sort it all out in Bitrix24 Open Channels. The **Bitrix24** channel routes those same messages to your AI Employee: it replies right in the Open Channel, conversations and cards stay where they are, and nobody has to learn a new interface. ## What you need * Bitrix24 portal administrator rights: only an administrator can access webhooks and Open Channel settings. * At least one channel connected in the Bitrix24 **Contact Center**, such as Telegram, WhatsApp, Avito, or website live chat. Your AI Employee hears the customer through the Open Channel, not directly. ## Connection Open **Applications → Developer resources → Other → Inbound webhook**. In the **Permissions** block, select: CRM (`crm`), Drive (`disk`), Chatbot creation and management (`imbot`), Chat and notifications (`im`), Users (`user`), Tasks (`task`), Open Channels (`imopenlines`). Creating an inbound webhook in Bitrix24: the URL field and the Permissions block with seven permissions selected Click **Create** and copy the **Webhook to call REST API**, an address like `https://your-portal.bitrix24.ru/rest/1/abcdef0123456/`. Open the AI Employee → **Channel** tab → **Connect New** → **Bitrix24**. Enter a **Channel Name** (only you see it), paste the webhook address into the **Webhook URL** field, and click **Create**. The Add Channel panel with the Bitrix24 platform: channel name, Webhook URL field, and message buffer settings The channel appears on the **Connected** tab, and the platform creates a chatbot in Bitrix24: you can see its number in the **bot\_id** field inside the channel. Until the chatbot is assigned as the responsible person, messages from the Open Channel don't reach your AI Employee. Open the **Contact Center** and select the channel where the AI Employee should reply, such as Telegram, WhatsApp, or website live chat. Bitrix24 Contact Center: channel tiles for telephony, website widget and form, live chat, WhatsApp, Telegram, VK In the channel settings, open the **Chatbots** section, turn on **Assign chatbot as responsible person when a customer makes contact**, and select the bot that appeared after you connected the channel. Open Channel settings in Bitrix24, Chatbots section: the checkbox for assigning the bot as responsible and the bot selection list | Field | What to set | | - | - | | **When to connect the chatbot** | At the start of every new conversation | | **When to transfer the conversation from the chatbot to the queue** | Don't transfer; otherwise Bitrix24 will take the conversation away from the AI Employee and hand it to the managers' queue | | **When to disconnect the chatbot** | After the conversation ends | Save the Open Channel settings. Open the channel by clicking it on the **Connected** tab and go through the settings: pipeline filter, operator involvement, message buffer, Co-Pilot mode. ## Test it Write to the connected channel as a customer, for example to the Telegram bot set up in this Open Channel. Your AI Employee replies in the same chat, and the conversation appears in [Chats](/en/guides/chats). ## Channel settings The defaults don't suit every portal, so go through the settings right away. **Pipeline Filter** narrows the scope. While nothing is selected, the AI Employee replies to all deals. Select one pipeline, and it works only there, leaving your cold leads or post-sale support alone. The Pipeline Filter block with the hint: if nothing is selected, the AI replies to all deals **Connect Operators in Parallel** adds the selected managers to the conversation after the AI Employee replies, so they can see the conversation and step in. The **Forced** toggle adds them on every incoming message without waiting for a reply. The Connect Operators in Parallel and Forced toggles in the channel settings **Message Buffer** makes the AI Employee wait before replying, since customers often split one thought into several messages. The default **Buffer Time** is 8 seconds; in Open Channels, set it longer, up to 120 seconds, because messages arrive with a delay. **History Auto-Cleanup** deletes the conversation on a schedule. Auto-cleanup deletes messages permanently. ## Co-Pilot mode The Co-Pilot Mode toggle: the AI replies with notes for the manager instead of messages to the customer With **Co-Pilot** on, the AI Employee sends nothing to the customer. Instead, it writes its reply as a note for the manager, who reads the suggestion, edits it as they see fit, and sends it themselves. This is the best way to roll out an AI Employee in a live sales team: for a week it writes notes, you review what it suggests and refine the [Persona](/en/guides/persona), and only then do you let it talk to customers. ## Actions in Bitrix24 The channel only brings in messages: it doesn't create deals or leads, set tasks, or assign responsible persons. The Bitrix24 talent does all that: [how to connect it](/en/guides/connect-talents#how-to-connect-bitrix24).