Appearance
Assistant configuration
Everything your sales assistant is — its role, tone, knowledge, tools, working hours, and handoff rules — is configured on one page: Agents in the sidebar, at app.mychatbot.app/configuration (the page is titled Agent configuration). This is the reference for every setting on that page: what it does, where it hides, and the nuances that aren't obvious from the labels.
Not the Agents Platform
This page covers the Sales Platform assistant — the AI that answers your customers. Building autonomous agents for open-ended work happens elsewhere; see the Agents Platform docs for that.

Page map
| Section | What you configure there |
|---|---|
| Header | Which assistant you're editing, Create copy, and the Overview / Editor toggle. |
| Status | One switch: is the assistant working or paused. |
| Profile | Name and Communication style (tone). |
| Toolkit (collapsible) | Knowledge base, Web Search, Video messages, Integrations, Lead Management, Contact Details, Client Context, Follow ups. |
| Settings (collapsible) | Work Hours, Flight Control, Handoff Control, Reply & Tools. |
| Role, Notes, Skills & Communication | Instruction cards: Role, Notes (live notes in force right now), one card per skill, + New Skill, Communication — plus Sub-Agents when present. |
| Test Dialog (right rail) | A live chat with the current configuration. |
Header: pick, copy, switch views
| Control | What it does |
|---|---|
| Assistant dropdown (Select an assistant) | Switches between assistants on the account. The URL carries the selection (/configuration?assistantId=…), so you can bookmark or share a direct link to one assistant's config. |
| Create copy | Duplicates the current assistant — all settings and its integrations are cloned so the copy is fully independent. You'll see "…has been cloned from …, now edit and save it". Disabled while you have unsaved instruction changes ("Save changes before duplicating the assistant"). |
| Overview / Editor | Toggles between the settings overview and the full-screen instructions editor. The page remembers which view (and which tab) you had open per assistant. |
Adding and removing assistants
Your first assistant is created during onboarding. After that, Create copy is the only way to add another one — the button's tooltip literally reads "Add a new assistant". There is no blank "new assistant" form: clone your closest existing assistant, rename it, and edit from there.
There's no delete button either. Assistants are hidden, not deleted, so chat history and data survive: go to Settings → Agent Actions (app.mychatbot.app/profile/agentActions), pick the assistant under Agent Visibility, and click Hide Agent — "This will not delete any chat history or data associated with the agent." Hidden assistants move to the Hidden Agents list on the same screen, where Restore brings one back.
How many assistants can be active?
You can create any number of assistants, but on the Standard and Pro plans only "1 active sales assistant at a time" works — the most recently updated stays active, the others switch off automatically. Multipro and above allow "multiple active sales assistants — no auto-deactivation". On a single-assistant plan the Channels page also redirects you to the active one: "Your subscription plan does not allow using multiple assistants."
Saving — and when changes take effect
Most of the page saves itself; a few cards want an explicit Save:
| Saves | Controls |
|---|---|
| Instantly, on change | Status, Name (on click-away), Communication style, the knowledge base pick and Strip knowledge base results from history, Web Search and Search model, Video messages, Lead Management, Contact Details, Client Context (switch and each field), Follow ups, the Static / Dynamic switch on skill cards, Flight Control control numbers (add, switch, access, remove), and Live notes (Add note, Remove). |
| On explicit Save | Work Hours (you'll see "Schedule saved"), Handoff Control, Reply & Tools, and all instruction text in the editor (Role, skills, Communication). |
Once saved, a change is live from the next message the assistant answers — including conversations already in progress, on every channel. There is no propagation delay and no "new chats only" rule: the assistant re-reads its configuration on each reply.
Status
A single switch with the label "Agent <name> is working" / "Agent <name> is paused". Paused means the assistant stops replying in every channel connected to it (which channels those are is decided on the Channels page — see which assistant answers which channel); chats keep arriving and operators can still answer.
Assistants can pause themselves
If your balance runs out or the subscription lapses, assistants are switched off automatically and the reason is recorded. A payment switches them back on (usually within the hour) — unless you had paused the assistant yourself: the platform never overrides the owner's switch, and flipping it in either direction marks the assistant as owner-managed from then on. Assistants muted before mid-August 2026 carry no recorded reason and must be switched on by hand. Turning an assistant on without any plan is refused with "active subscription required to enable agent." A Work Hours schedule flips this same switch on and off automatically. Full rules: Balance & billing → How assistants come back.
Which assistant answers which channel
Pointing a channel at an assistant does not happen on this page — it happens on the Channels page (app.mychatbot.app/channels). The same assistant dropdown sits at the top there; the title becomes "Channels for <assistant>", and every channel you connect while that assistant is selected belongs to it.
- To split channels between assistants, select assistant A in the dropdown and connect its channels, then switch the dropdown to assistant B and connect the rest.
- A channel identity connects to one assistant at a time — Telegram, for example, warns: "This bot token is already connected to another assistant. Please revoke it first or use a different bot." To move a channel, disconnect it under its current assistant, switch the dropdown, and reconnect it under the new one.
- Each channel has its own Connect agent to this channel switch, separate from the assistant's Status. Saving or updating a channel's credentials turns that switch off — re-enable the channel after every credential edit, or the assistant goes silent there while its Status still reads "working".
Profile
"Let's give your agent a personality."
| Setting | Details |
|---|---|
| Name | Up to 50 characters. Saves as soon as you click away from the field. |
| Communication style | A dropdown of tone presets, applied on top of your instructions. |
The Communication style options:
| Style | Use when |
|---|---|
| Short and concise answers | Fast-paced messenger chats, mobile-first audiences. |
| Normal | Default, balanced replies. |
| Verbose and descriptive answers | Complex products that need explanation. |
| Energetic tone that spotlights offers and ends with a clear call to action. | Promo-driven selling. |
| Warm, empathetic tone that acknowledges concerns and guides calmly to a resolution or escalation. | Support-heavy or sensitive niches. |
| Trustworthy advisor tone that explains options plainly and suggests gentle upsells when relevant. | Considered purchases, consultative sales. |
Where's the welcome message?
The first-message greeting is not on this page. Set it per channel under Channels → your channel → Instant welcome message (it's also asked during onboarding). That way each channel can greet differently.
Toolkit
"Provide your assistant with business knowledge and tools to handle product data, process leads, and close deals."
Click the Toolkit header to expand it.
Automations live in Custom MCP
Automatic calls to your MCP server (sending conversations, client-reply notifications) are configured in Custom MCP.
Knowledge
Assigns a knowledge base to this assistant and shows which sources feed it.
| Control | What it does |
|---|---|
| Knowledge base dropdown | Pick which knowledge base this assistant reads (Default, Knowledge base №2, …). Create new jumps to the Knowledge page. If none exists yet you'll see "No Knowledge base added yet" with a Connect knowledge base button. |
| Source cards | The business and product sources connected to the selected knowledge base, with shortcuts into the relevant Knowledge page section. |
| Strip knowledge base results from history | Removes knowledge-base results from conversation history before the next AI call. Reduces input tokens for faster, cheaper replies. The assistant can re-search if needed, and full results are always saved in chat logs. |
Nuances of the strip toggle:
- It applies to all sources in the selected knowledge base at once — one switch, every source.
- Calls channels take a snapshot of the knowledge base when the channel is created — later changes to the knowledge base reach a voice assistant only after the Calls channel is recreated.
- Turn it on for large catalogs where every reply triggers a product search: history stays lean and costs drop. Leave it off if your flow depends on the assistant re-reading earlier search results without searching again.
Web Search
"Real-time internet search for improved responses."
| Control | What it does |
|---|---|
| Web Search switch | Lets the assistant search the live internet when your knowledge base doesn't cover a question. |
| Search model | Sonar (default) or Sonar Pro Search. The Pro option digs deeper, but "each Sonar Pro search call counts as extra message usage". |
Video messages
"Recognizes speech in videos clients send, so the assistant replies to what they said — just like voice messages."
When a client sends a video, the assistant reads what was said in it and replies to that. A client who films a damaged item and describes the problem is answered as a complaint — and if your instructions tell the assistant to pass complaints to a manager, it hands the chat over (see Handoff Control for what happens next).
| Control | What it does |
|---|---|
| Video messages switch | On by default. Turn it off and the assistant no longer gets the words from videos; your team can still play every video in the chat. Telegram's round video messages are always transcribed, like voice messages. |
Nuances:
- The words are added to the same message as the video, starting with "Speech in the video:" — your team sees exactly what the assistant read. A video with no speech is marked "[The video has no speech]", so the assistant knows a silent video arrived and can ask the client to describe it.
- The assistant hears the video; it does not watch it. Anything shown but not said is not passed on. One exception: when a client replies to your Instagram story video and Instagram won't return the story itself (for example while Instagram limits how often it can be read), the platform takes one still frame of the video and shows it to the assistant as an image. The frame is also saved in the chat. If no frame can be taken, the message carries the note "[You cannot see what this story shows: its content could not be loaded. Ask the client which item they mean, or hand the chat to a manager. Do not assume it is a product from earlier messages.]" (see Instagram).
- Words are added only when the assistant answers. In a chat handed to your team, while the assistant is paused (including outside Work Hours), or for a video over the limits below, the video arrives without them.
- Works on Instagram, Facebook Messenger, Personal WhatsApp, Telegram (bot and Telegram Account), Viber (bot and E-Chat Viber) and TikTok. Videos up to 25 MB (Telegram bots receive files up to 20 MB); up to three videos at a time.
- The assistant replies in text, even with Voice reply on — a video is not treated as a voice message.
- A reply to a video can take a few seconds longer: the speech is recognized first.
- No extra charge: recognizing the speech is free; the assistant's reply costs one message as usual (see what is charged).
Integrations
"Connect to external tools and CRM or manage leads locally."
Read-only status cards for the integrations attached to this assistant, plus a Connect integrations link that opens the Integrations page pre-scoped to this assistant. If nothing is connected you'll see "No integrations connected".
Lead Management
Controls how this assistant's conversations turn into leads.
| Setting | What it does |
|---|---|
| Manage leads locally | Covers funnel statuses only. On: pipeline statuses are created and edited inside MyChatBot, and the assistant can set them. Off with a CRM connected: funnel statuses are synced from your CRM and shown read-only here. Labels are controlled separately by The agent sets labels. |
| The agent sets labels | On by default, and a missing value counts as on whether or not Manage leads locally is on. Turn it off to take the assistant's label tools (set_client_label, remove_client_label) away. It sits beside the Labels card. |
| Disable auto-assign pipeline | By default a new assistant is automatically attached to your default pipeline. Turn this on to stop that and pick the pipeline yourself. |
| Pipeline selector | Which pipeline this assistant files its leads into. Statuses shown below are filtered to this pipeline. |
| Pipeline Statuses / Labels cards | The status and label sets the assistant can apply. Pipeline Statuses is editable (edit/delete) when managing locally and read-only when synced from CRM. Labels is always editable. |
Statuses and labels feed the instructions editor
When the assistant can set statuses and labels, your exact status and label names get highlighted inside the instructions editor — reference them by name in your sales script ("move the client to Qualified") and the editor shows you they match.
Contact Details
"Allow the agent to save client name, phone number, and email when shared during conversation."
One switch. When enabled, the assistant can update the client's Full name, Phone number, and Email address on the client card whenever the customer shares them in chat — no instruction writing needed.
With the switch on or off, the assistant also saves contact details when it notifies an operator or hands the chat to a human: the name, phone and email it passes along go onto the chat's own client card, but only into fields that are still empty. That path never overwrites a value already on the card. It also skips a phone number that another client in the account already has, and does not merge the two clients. If the save fails, the notification or handoff still goes ahead. Turn Contact Details on when the assistant should keep these fields up to date at any point in the chat, including correcting a value already saved.
Client Context
"Collect and maintain client data points across conversations. Data is shared across all channels for the same client."
Contact Details covers name/phone/email; Client Context is for everything else you want remembered per client — budget, size, deadline, preferences.
| Control | What it does |
|---|---|
| Client Context switch | Enables collection. Behind the scenes this adds an always-active skill to the assistant (you'll see it as a card under Role, Notes, Skills & Communication marked Always active). |
| Data points to collect | The schema of what to gather. "Define what information the agent should collect from clients during conversations." |
| Add field | Adds a data point: a Label (e.g. Budget Range) and a Key (e.g. budget_range — generated automatically from the label, lowercase with underscores; you can override it). |
| + Add description | Optional per-field hint for the AI (e.g. "The client's budget for the project") — use it when the label alone is ambiguous. |
Collected values persist across conversations and channels: the same client on Instagram and WhatsApp shares one context.
Follow ups
Attach follow-up sequences to this assistant.
- Pick from your existing follow-ups or Add follow-up to create one; Open Follow-Ups jumps to the editor.
- Drafts don't run: "Draft follow-ups won't work until published. Open the Follow-Ups editor to publish." Draft entries are chip-marked Draft so you can spot them.
- Clear all detaches every follow-up from this assistant.
Settings
Click the Settings header to expand four cards: Work Hours, Flight Control, Handoff Control, and Reply & Tools.
Work Hours
Defines when the assistant is on duty.
| Control | Details |
|---|---|
| Enable working time | Master switch for the schedule. Off = always on duty. |
| Day selector + Time-from / Time-to | Per-day working window. Apply this schedule to all days copies the current day everywhere. |
| Time zone | "Select from the list or start typing. The agent uses this timezone as its current time — set it even if you don't limit working hours." The schedule is interpreted in this zone, and it is the clock the assistant quotes dates and times from in every conversation. Live notes end on this clock too, and days_until (see Agent Tools below) counts from today in this zone. With no zone set, the assistant uses UTC. |
| Breaks (time off) | Windows inside working hours when the assistant goes quiet (lunch break). Add break for more than one. |
| Extra working hours | Windows outside the main schedule when it should still work. Add extra hours to stack several. |
You'll see "Schedule saved" on success. If a field is invalid, nothing is saved: you see "Check the fields with errors and save again" and the day with the error opens. If saving fails, you see an error ("Failed to save schedule" by default) and the window stays open with your changes, so you can try again.
Outside the schedule, the assistant's Status switch flips to "Agent <name> is paused" automatically — exactly as if you'd paused it by hand. From the customer's side that means silence, not an away-reply: off-hours messages land in the chat unanswered and stay visible to your team, and operators can reply manually the whole time. When hours resume the switch flips back on and the assistant answers messages that arrive from then on — it does not go back and reply to the off-hours backlog, so have an operator sweep those chats (or rely on Follow ups to re-engage).
Flight Control
"Get information and give tasks to your agent."
Flight Control is where your business and its assistant talk to each other. The Flight Control card opens a window with two parts:
| Part | What it does | Plan |
|---|---|---|
| Telegram group | The assistant posts updates for your team to one Telegram group: handoffs, alerts, new leads. | Every plan |
| Email alerts | Optional, and separate from the Telegram group: the assistant's alerts also go by email to up to three addresses you add. | Every plan |
| Control numbers | The people who run the business, such as the owner or a manager, give the assistant instructions and ask it questions from their own WhatsApp. | Pro and above |
The card sums up what is set, for example "Telegram connected · Email alerts · 2 numbers", "1 number", or Not set.
Telegram group
- Click Open in Telegram (or Copy Link & Close) to invite @mcbassistant_bot to your team's Telegram group. Telegram guides you through picking the group.
- "Once the bot joins the group, it will link the chat to this assistant automatically."
- Already linked? The block shows Already connected with the group. "Want to connect a different group? Disconnect first, then invite the bot again."
One group per assistant; Disconnect frees it. While a group is linked, Client message edit alerts can also "post a message to this group when a client edits a message they already sent."
The group is for updates: the assistant doesn't act on what people write in it. What it posts there is listed under Notifications → Telegram. To give the assistant instructions, use a control number.
Email alerts
An optional way to get the assistant's alerts by email, on its own or beside the Telegram group. It is off until you turn it on, and it doesn't depend on Telegram: with Send alerts by email on, every alert the assistant raises (the operator ping, see Notifications → Telegram) goes to your addresses whether or not a group is connected, and the Telegram group gets its message as before.
- Turn on Send alerts by email.
- Add up to three addresses. Add my email fills in the email you signed in with; nothing is added until you tap it.
- Each new address gets one confirmation email and nothing else until whoever reads it confirms. The list shows each address as Confirmed, Waiting for confirmation (with Send confirmation to send it again), or Unsubscribed. Your account's own email needs no confirmation.
What the email carries: the assistant's message, the client's name, phone and email, the channel, and a link to the chat. Replies to it don't reach the client; open the chat to answer. Links the client or the assistant wrote are shown as text, not as links.
| Limit | What happens |
|---|---|
| One email per chat per hour | Later alerts in the same chat wait and arrive together in one email |
| More than 10 alerts in an hour | The rest of that hour arrives as one summary email at the top of the next hour |
| 100 alert emails per account a day, 40 per address | The rest of the day's alerts are not emailed (Telegram is not limited) |
| 2 confirmation emails per address a day, 10 per account | Send confirmation says so; try again the next day |
Every alert email has a one-click Stop them for this address link. It stops only that address, only for this assistant, and adding the address again doesn't undo it: the address gets alerts again only after its reader confirms a new confirmation email.
Control numbers
A control number is the phone of someone who runs the business: the owner, a shift manager, the kitchen. When that person writes or sends a voice note to the assistant's WhatsApp number, the assistant doesn't treat it as a customer. It takes the message as an instruction for its customer chats, confirms it, and applies it.
| The person sends (text or voice) | What the assistant does |
|---|---|
| "The burger is sold out today" | Saves a live note that ends at midnight and confirms in one line, with the end time and time zone, for example "Noted: no burgers until midnight (Europe/Vienna). 2 notes active." Customers who ask for a burger are told it's sold out. |
| "No terrace for the next two hours" | Saves a note that ends in two hours. |
| "We close at 21:00 tonight" | Saves a note until the end of the day. Customers asking about opening hours tonight hear 21:00. |
| "Which notes are active?" | Lists the notes in force. |
| "Cancel the burger note" | Cancels that note and says which one. If more than one note could be meant, it asks which first. |
How the assistant handles these messages:
- Each instruction becomes a live note. The assistant rewrites it so it stands on its own, for example "The burger is sold out today; suggest another dish.", one note per instruction, in the language the person wrote in.
- A note lasts until it is cancelled, unless the person says when it ends: "for two hours" (1 minute to 7 days), "today" or "tonight" (midnight), "until tomorrow" (the end of tomorrow), "until closing" (today's closing time from Work Hours), or a date and time, "until 18:00 on Sunday". These are the same choices as Ends in the dashboard, on the same clock: see When a note ends. If no zone comes from Work Hours or the assistant's WhatsApp number, the control chat goes by the country of the control number before falling back to UTC.
- Every customer chat applies the notes, silently. From the next reply on, the notes reach every customer conversation of this assistant on every channel, including the follow-ups and outreach it writes. The assistant never quotes or lists notes to customers and never says who sent them.
- Voice notes work like text. If a recording is unclear, the assistant asks before it saves anything. Photos can't be kept as notes.
- It is not a customer chat. The assistant answers briefly, in the person's language. It doesn't sell to them, take their order or hand the chat to an operator. Asked for something it can't do here, it says what it can do.
- It works while the assistant is paused, for example outside Work Hours. The WhatsApp channel itself must be switched on.
How the control chat works
In a control chat the assistant keeps its own instructions, with control-chat rules around them. Before each answer it knows:
- Who is writing. Someone who runs the business, by the name on the list, never a customer.
- The business. It reads its Role and its Static skills in full, and its Dynamic skills by name, to learn the menu, services, prices, policies and hours. That way a note fits the business, and it can answer questions such as "What do your instructions say about delivery?". The customer rules in them don't apply here, and neither does any tool they name that this number hasn't been given. Communication style isn't used in this chat.
- Its tools. The live-note tools, plus what you set this number to View. See Tools in the control chat.
- The live notes and the clock. The notes in force, the current time, the time zone and where it comes from, and today's closing time when Work Hours have one.
Add a number
- Open Agents → your assistant → Settings → Flight Control.
- Under Control numbers, click Add number.
- Enter a Name (for example Owner, up to 60 characters).
- Under Phone number, pick the country from the flag list, then type the number ("Choose the country, then type the number."). The list starts on the country of the assistant's Personal WhatsApp number, or on International when it has none; you can also type the full number with
+and the country code. - Click Add. It stays off until the number is a real one for that country. A wrong number shows "This is not a valid phone number. Check the country and the digits, e.g. +43 660 0000000.", and one already on the list shows "This number is already on the list." The number takes instructions from its next message.
Add the phone the person writes from, not the business number the assistant answers on. Control numbers belong to one assistant: to steer two assistants, add the phone to each.
Each number on the list shows its name, phone and what it can ask for, with two controls:
- The switch, Take instructions from <name>. Off keeps the number on the list, but its messages are handled like any customer's.
- The bin icon, Remove <name>, which asks once: "Remove <name>? This number won't be able to give instructions anymore." Notes it added stay until they end or you remove them under Live notes.
Where instructions are accepted
The block lists each WhatsApp number of the assistant, with the WhatsApp Business logo for WhatsApp and the WhatsApp logo for Personal WhatsApp, and a line saying whether it takes instructions. A number that can't is greyed out. With two numbers of the same kind, each row also shows its number.
| Channel | Takes instructions? |
|---|---|
| WhatsApp (WhatsApp Business, including a number that stays on the WhatsApp Business app) | Yes. |
| Personal WhatsApp | Yes. If the line under the number says "This number can't take instructions yet. Contact support and we'll switch it on.", contact support. |
| Telegram, Instagram, Messenger, the website chat, calls and every other channel | No. |
| Group chats | No. Only one-to-one chats count. |
With no WhatsApp number on the assistant you'll see "Connect a WhatsApp number to this assistant first. Control numbers send their instructions to that number." A channel that is off reads "This channel is switched off."
Messages sent from the business number itself, such as replies your team types in the WhatsApp Business app, are operator replies in that customer's chat, never instructions. A control number is a different phone that writes to the business number.
What this number can ask for
Every control number can add, remove and list live notes (Live notes, Always on). On top of that, you can let a number look things up for you. Click What this number can ask for under the number (it opens by itself right after you add one) and set each group to Off or View. "View lets the number look things up. It cannot change anything." A new number starts with every group off, shown as Live notes only.
| Group | What the number can ask about | For example |
|---|---|---|
| Clients and chats | Clients, their chats, notes and tasks | "What did <customer name> ask yesterday?", "Who asked about catering this week?" |
| Orders | Orders and order totals | "How many orders today?" |
| Bookings | Bookings, staff and services | "Which tables are booked for tonight?" |
| Assistant setup | The assistant's instructions and FAQ | "What does the FAQ say about parking?" |
| Usage | AI usage, orders, active chats and new clients over a period | "How many chats did we have this week?" |
Answers come from this account's data only. From WhatsApp, a control number can't change client records, send messages to customers or edit the assistant's instructions; do that in the dashboard. A control number can't see the balance either: check it under Balance & billing.
These are the tools the assistant can use in a control chat. The live-note tools are always there. Each group adds its tools only when the number has it set to View.
| Group | Tool | What it does |
|---|---|---|
| Live notes (always on) | add_live_note | Saves an instruction as a note, with when it ends. The same instruction twice is one note. |
list_live_notes | Lists the notes in force, with who added them and when they end. | |
cancel_live_note | Cancels one note in force. | |
| Clients and chats | list_clients | Finds clients by name, phone or email, label, status or pipeline. |
scan_clients | Goes through every client, page by page, for questions about all of them. | |
get_client | One client's card: contact details, status, labels and what the assistant collected. | |
client_list_notes | A client's notes. | |
client_get_note | One of those notes in full. | |
client_list_tasks | A client's tasks. | |
list_chats | Recent chats, newest first: channel, last activity, and whether an operator is needed. | |
get_chat | One chat's details: channel, client and follow-up state. | |
get_chat_messages | What was said in a chat, including call transcripts. | |
list_labels | The account's labels. | |
list_pipelines | The account's pipelines. | |
| Orders | list_orders | Orders, by date, client, status or source. |
get_order | One order in full, by its number. | |
get_order_stats | Order count and revenue for a period, by channel and status, plus how many won deals still have no amount. | |
| Bookings | list_events | Bookings, by date, status, staff or service. |
get_event | One booking in full. | |
get_event_stats | Booking counts for a period, by status and staff. | |
list_calendar_staff | The staff set up for bookings. | |
list_calendar_services | The services set up for bookings. | |
| Assistant setup | get_assistant | The assistant's settings, instructions and skills. |
list_faq_knowledge_base_entries | The questions and answers in the FAQ knowledge base, and the text of connected Google Docs. | |
| Usage | get_usage_summary | AI tokens, orders, active chats and new clients for a period. |
get_token_usage | AI tokens for a period, for the account or one assistant. |
Apart from the live-note tools, every tool here only reads. Levels that change data, such as updating a client, sending a customer a message or editing the FAQ, aren't offered yet, so the dashboard shows only Off and View. Deleting, billing, channels, exports and automations never reach a control chat. For every parameter of these tools, see the Sales tools reference.
Plan, cost and limits
Control numbers come with the Pro plan and above. On a lower plan the block shows a PRO chip, "Control numbers are available on the Pro plan and higher." and a See plans link. If your plan later drops below Pro, the numbers already there stay and you can still remove them; adding a new one needs Pro again. Each reply the assistant sends to a control number uses your account's messages, like any other reply (see Balance & billing).
| Limit | Value |
|---|---|
| Control numbers per assistant | 5 ("You can add up to 5 numbers.") |
| Live notes in force per assistant | 50 |
| Characters per note | 500 |
| Name of a control number | 60 characters |
Add only phones you trust
What a control number sends changes what the assistant tells every customer, and View access lets it read your clients' chats, orders and bookings. Only messages that WhatsApp proves came from that number count; a group chat or a hidden number never does. Every note added or cancelled from a phone is also posted to the Flight Control Telegram group, when one is linked.
In Chats
The conversation with a control number shows up in Chats like any other, with a Flight Control badge next to the name: "Control number: messages here are instructions for the assistant, not a customer chat." It also gets a client card, like any chat.
Testing as a customer
A message from a control number always gets the control chat. To try the assistant the way a customer sees it, write from another phone, or switch Take instructions from <name> off for a moment. To check how the assistant applies a note, ask about it in the Test Dialog: live notes apply there too.
If a control number gets a customer reply
| Check | Fix |
|---|---|
| The number's switch is off | Turn Take instructions from <name> on. |
| The number on the list is not the phone the person writes from (another SIM, a wrong country code) | Remove it and add the right one. |
| They wrote to another assistant's number, or on a channel other than WhatsApp | Control numbers work only on this assistant's WhatsApp and Personal WhatsApp numbers. |
| The line under the channel doesn't say Takes instructions | Personal WhatsApp: contact support. A channel that is off: turn Connect agent to this channel on under Channels. |
| They wrote in a group | Only one-to-one chats count. |
Handoff Control
Rules for switching a chat between AI and a human operator.
| Setting | What it does |
|---|---|
| Stop word for manual handoff | A keyword under Handoff Mode. "When an operator types "<keyword>", the chat switches to operator mode." Typed in a chat already in operator mode, the keyword switches it back to AI. Other messages operators send from the channel's own app (their phone, the native inbox) don't interrupt the AI. A reply sent from Active chats is different: it always switches a chat the AI is handling to operator mode. |
| Implicit operator handoff | "Any message from an operator immediately switches the chat to operator mode." No keyword needed, including for messages sent from the channel's own app. Replies sent from Active chats switch the chat whether this is on or off. |
| Silent handoff (under Client Notification) | "Don't notify the client when switching between AI and operator." The customer sees one seamless conversation. |
| Replies from the WhatsApp Business app | For a WhatsApp number that stays on the WhatsApp Business app. "When you reply from the WhatsApp Business app in a chat the AI is answering, the AI pauses there. It takes the chat back after 10 minutes with no new message from you or the customer, and answers the customer if they're waiting. A chat that's already with a person stays as it is, and your reply takes it out of Needs attention; spam chats are left alone." Let the AI take the chat back is ticked by default; Minutes with no new message is "Leave empty for 10 minutes. From 1 to 1440 (24 hours)." Each reply from the app starts the minutes again, and so does each new message from the customer. The app's own greeting and away messages, notices a store system sends from your number to many customers (sign-in codes, cart reminders, order updates), and reactions don't pause the AI. If the AI has nothing to add for the customer when it takes the chat back, it stays quiet. Unticked, the chat stays with you like any other handoff. The AI comes back only when its AI switch is turned back on in the chat, or through Switch back to AI, a lead status set to Automatically enable agent on this status, a follow-up scheduled for the chat, or Auto return from handoff. This applies only to chats the AI is answering when you reply. A chat already with a person (handed over by the assistant or a stop word, taken in the dashboard, switched off by hand, or spam) keeps its handoff: your reply starts no return, and Auto return from handoff applies as before; your reply takes it out of Needs attention, as a dashboard reply does — see WhatsApp. With a stop word set, the stop word decides instead. The return is recorded in the chat's activity log. |
| Auto return from handoff (under Auto Return) | "If an operator disables the agent in a specific chat and doesn't re-enable it within 24 hours, the agent will be enabled automatically." |
| Keep chats switched off by hand in operator mode | "Auto return skips chats where someone turned AI off by hand. They stay in operator mode until AI is turned back on." "By hand" means the chat's AI switch, or a reply sent from the dashboard while Implicit operator handoff is off. Chats the assistant hands over, and chats switched by Implicit operator handoff or the stop word, still return. Unticked, chats switched off by hand return after the interval too. |
| Custom return interval (minutes) | Overrides the 24 h. "Leave empty to use the default (1440 min = 24 h). Min: 1, max: 10080 (1 week)." An automatic return is recorded in the chat's activity log. |
| Outreach failure alerts (under Owner Notification) | "Notify me on Telegram when an outreach can't reach a contact — they blocked messages, restricted their account, aren't on WhatsApp, or have no reachable number." Goes to the Flight Control group. |
| Automatic manager assignment | Assign a manager when a chat is handed over, the picking mode (By workload / In turn (random)) and the pool of eligible teammates — see Team & access → Automatic manager assignment. |
Keyword and implicit modes are mutually exclusive
The keyword field is "disabled while implicit handoff is enabled", and the implicit switch is "disabled while a keyword is set — clear the keyword to enable implicit mode." Pick one.
When the assistant hands a chat over and passes the client's name, phone or email, those go onto the chat's client card, into empty fields only (see Contact Details).
Changes here need an explicit Save.
Auto return settings apply to chats already in operator mode
Auto return counts from the moment a chat went to operator mode, or from an operator's last message in it, not from when you saved the setting. Auto return checks chats every minute, so a save that makes more chats eligible returns all of them to AI within about a minute:
- ticking Auto return from handoff;
- shortening or clearing Custom return interval (minutes) (empty means 24 h);
- unticking Keep chats switched off by hand in operator mode. Chats switched off by hand days ago come back at once, not after another interval.
Before such a change, open the Operator tab and note the chats that must stay with a person, so you can turn their AI switch off again right after saving. Turning Auto return from handoff off later does not take back chats it has already returned: each one stays in AI mode until its AI switch is turned off.
Reply & Tools
A modal with three groups of switches. The button below the other Settings cards shows Configured once anything in it is on, Not set otherwise. Changes apply after Save.
Language Controls
| Setting | What it does |
|---|---|
| Primary reply language | Force a default reply language (about 50 to choose from), or "keep it not set to follow the customer's lead." |
| Allow agent to switch chat language | Lets the assistant call the set_chat_language tool when needed — e.g. when the customer switches languages mid-chat. |
Agent Tools
| Setting | What it does |
|---|---|
| Allow agent to self schedule follow-ups | The assistant can call the schedule_message tool to queue its own reminders ("I'll check back tomorrow") without a pre-built follow-up sequence. |
| Hide calculus expressions from user | Uses the perform_calculus tool to keep computations internal — customers see the result, not the arithmetic. With this on, the assistant also gets the days_until tool: it counts the days and weeks from today, in the assistant's Time zone, to a date given as YYYY-MM-DD or DD.MM.YYYY with the year spelled out (2026-12-12 or 12.12.2026). A past date comes back as a negative day count with in_past set to true. Impossible or ambiguous dates, such as 31.02.2026 or a slash date like 12/12/2026, are refused, never guessed. Use it for rules that depend on how far away a date is, for example "hand the chat to a manager when the event is fewer than 13 weeks away". |
| User can have last word in a chat | Enables the end_conversation tool so the assistant can gracefully stop replying when the exchange is done (a final "thanks!" doesn't trigger yet another AI answer). |
Channel Representation — how replies are rendered in messengers:
| Setting | What it does |
|---|---|
| Remove Image Links | "Images are sent as native channel objects without the source link." |
| Remove Video Links | "Video links (YouTube etc.) are removed from the reply." |
| Remove File Links | "File links (PDF, Docx, etc.) are removed from the reply." |
| Send docs.google.com/document/… as attachments | Google Docs document links become real attachments. |
| Send drive.google.com/file/d/… as attachments | Google Drive file links become real attachments. The file must be shared as Anyone with the link → Viewer. Otherwise it can't be downloaded and reaches the customer as a link. On email, the chat also records the file as not delivered, along with the sharing fix. |
| Send links as separate messages | Telegram, Instagram and Messenger: each URL is sent in its own message. On Telegram this keeps the text from breaking on stray Markdown characters; on Instagram and Messenger every link gets its own preview card (only the first link in a message is previewed). On other channels links stay inline. |
| Split reply by paragraphs | Each paragraph of a long reply is sent as its own message, on chat channels. Very short paragraphs are joined to the one before them, and one reply is capped at a few messages. Voice channels, email, SMS and the WordPress/OpenCart widgets are unaffected — they always get the reply as one piece. |
| Tappable options on WhatsApp | On WhatsApp Business, when a customer is to pick one of a few options, the assistant shows them as buttons (up to 3) or a list (up to 10), so the customer taps instead of typing. The chat history shows the options as a numbered list. Not on Personal WhatsApp or other channels. See Buttons and lists in replies. |
On email, Remove Image Links, Remove Video Links and Remove File Links remove a link only once its file has actually been attached. A file that fails to attach keeps its link where the reply put it. So does a link the email doesn't try to attach, such as a Google Drive or Google Docs link while its Send … as attachments switch is off. When an attached file's link was the only thing on a bullet line, the empty bullet line is dropped. See Email: Sending replies.
Role, Notes, Skills & Communication
The instruction cards. On a brand-new assistant you'll first see two starter cards — Businesses offering products and Businesses providing services — that route you into initial instruction setup. Once instructions exist, the cards become:
| Card | Contains |
|---|---|
| Role | The core instruction: who the assistant is, your sales script, what it must do. Answering questions is assumed by default. |
| Notes | The live notes in force right now — "2 active", or "No notes yet". Edit opens them in the editor. |
| Skill cards (one per skill) | Named instruction modules you split out of the role (returns policy, booking flow, objection handling). |
| + New Skill | Creates a new skill and opens it in the editor. |
| Communication | Communication guidelines — "the communication guidelines your agent follows when talking to customers." Tone and formatting rules, separate from the sales script. |
Clicking Role or Communication opens a review modal ("Only verified instructions are displayed") with:
- Edit instruction / Edit communication guidelines — after a confirmation ("Any changes you make will overwrite the current configuration"), opens the editor.
- See Connected Tools — "the signatures of the connected tools; use them to adjust behaviour within instructions and tune what values should be passed if needed" — every tool with its Parameters and which are (required).
Static vs Dynamic skills
Each skill card carries a switch:
| Mode | Meaning |
|---|---|
| Static | Always included in the assistant's system prompt — always "in mind". |
| Dynamic | Loaded by the assistant on demand via a tool call — cheaper, fetched only when relevant. |
Keep skills the assistant needs on every reply Static; push long, occasionally-needed material Dynamic. The Client Context skill is system-managed and shows Always active instead of the switch.
Sub-Agents
If some of your integrations spawn helper agents, a Sub-Agents block appears below the cards — "Sub-agents are created automatically based on internal integration settings." You can Edit each sub-agent's instructions; its own tool list shows in the editor's Tools panel while you do.
Live notes
"Short instructions your assistant follows in every customer chat, on top of its instructions, until you remove them or they expire."
Live notes are the Notes card in Role, Notes, Skills & Communication, right after Role, so everything that shapes the assistant's replies is in one place. The card shows how many are in force; Edit opens Notes in the instructions editor (it's also in the editor's section list, right under Role). There you see the notes in force, whether a control number sent them or someone typed them here. Each note shows its text, who added it (the control number's name, Control number or Dashboard), when, and Until <date> or Until removed, with a Remove button. These times are on the assistant's clock (see When a note ends). A note that expires drops off the list by itself.
To add a note:
- Under Add note, type the instruction, for example "The burger is sold out today". Up to 500 characters; the counter shows how many you've used.
- Pick when it ends under Ends. Until removed is the default.
- Click Add note (or press Ctrl+Enter, ⌘+Enter on a Mac).
A note applies from the next reply, in every customer chat of the assistant on every channel, and in the Test Dialog. Remove ends it straight away. With no notes you'll see "No live notes right now. Add one below, or send one from a control number (Settings → Flight Control)." With 50 in force the form is locked: "You can have up to 50 live notes. Remove one to add another."
Write a note the way you'd brief a colleague: one instruction, plain words, and what to offer instead. Notes take priority over older information, so use them for what changes today or this week, and put lasting rules in the Role or a skill.
Live notes come with the Pro plan and above. On a lower plan the Notes card and the notes in the editor show a PRO chip, "Live notes are available on the Pro plan and higher." and a See plans link; notes already there can still be removed.
When a note ends
| Ends | The note ends |
|---|---|
| Until removed | Never by itself. It stays until you or a control number remove it. |
| In 1 hour / In 3 hours | One or three hours after you add it. |
| End of today | At midnight tonight. |
| Closing time (21:00) | At today's closing time from Work Hours, shown in brackets. Offered only while it is still ahead. |
| End of tomorrow | At midnight at the end of tomorrow. |
| Pick date & time… | At the date and time you pick. It must be ahead. |
All these times are in the assistant's time zone, and the form names it, for example "Times are in Europe/Vienna (from Work Hours)." The zone comes from:
- the Time zone under Work Hours, even when working hours are off;
- else the country of the assistant's WhatsApp number ("Times are in Europe/Vienna (from your WhatsApp number).");
- else UTC ("Times are in UTC — set a time zone in Settings → Work Hours.").
So a note set to end at 18:00 ends at 18:00 where the business is, whoever adds it and wherever their computer is. Set the Work Hours time zone if the assistant's clock is wrong.
- Closing time is when today's last working stretch in Work Hours ends. Extra working hours after the main window make it later, a break that runs to the end of the day makes it earlier, a lunch break doesn't change it, and a shift past midnight closes the next morning. It's missing when no working time is left today, and it drops off the list if it passes while the form is open ("Today's closing time has passed. Pick another end time."). From a control number, "until closing" with no closing time left makes the assistant ask for a time instead.
- A time skipped by a clock change can't be picked. On the night the clocks go forward, a time in the missing hour shows "That time doesn't exist in Europe/Vienna (the clocks jump forward then). Pick another time." and Add note stays off.
- A picked time that has already passed shows "Pick a time that is still ahead."
The instructions editor
Open it with any Edit button or the Editor toggle in the header. The section list holds Role, Notes, each skill, and Communication; Role, skills and Communication open a full markdown editor. Notes opens the live notes list and the Add note form instead: notes are saved as you add or remove them, so Save and Reset don't apply there.
| Feature | Details |
|---|---|
| Tool highlighting | Names of connected tools light up inside the text, so you can see at a glance which tools your script actually references. Your pipeline status and label names highlight too (when the matching tools are connected). |
| Save / Reset / Close | Reset returns to the last saved text. Leaving with unsaved changes prompts "You have unsaved changes. Are you sure you want to leave?" |
| History | Every saved version of the instruction, marked Current / Latest / Oldest, with Restore. Restoring loads the old text as a draft diff — review it, then Save to make it current. |
| Inline autocomplete | AI suggestions while you type. Turn off (along with editor font and line numbers) in Settings → General Settings. |
Side panels
Four icon buttons switch the right-hand panel:
| Panel | What it's for |
|---|---|
| Tools (Available Tools) | "Tools referenced in Role & Sales Script are highlighted. Reference their exact names when you want the agent to call them." Click a tool to jump to where it's mentioned. Each tool has a Test button — fill in its parameters, Run, and inspect the Result without leaving the editor. |
| Context (Context Knowledge) | "Preview which knowledge snippets <assistant> will read before replying." Type a client message ("Ask about pricing, catalog, docs...") and see exactly which entries come back, from which knowledge base, and how many words each adds. |
| Test dialog | The live test chat, right next to the text you're editing. |
| Script lab (Role & Skills Builder) | An AI co-writer: "Use it to build role, character and skill set for this agent." Describe what you want in plain language; it drafts or edits the Role and skills, and the editor shows the proposed changes as a diff for you to save or reset. Only available on the Role & Skills tab. |
Tool Test runs are (mostly) live
Test → Run executes the actual tool once, with no AI reply around it — so it doesn't consume an AI message, but the tool does its real job: searches return live data, a Web Search run counts usage like any search, and write-type integration tools hit the live connected system. The exceptions are sandboxed: pipeline status and label tools, operator handoff/notification tools, and the client-context save tool only return a simulated confirmation in test mode.
Test Dialog
The chat on the right side of the page (a swipe-in drawer on mobile) talks to the assistant with its current saved configuration — knowledge base, tools, style, everything. Clear chat wipes the test conversation so you can rerun a scenario from a clean slate.
Test replies are free
Test Dialog replies are generated by the same engine as customer replies but are recorded as test usage and do not draw on your message balance — rehearse as long as you like. Because they are not billed they also do not appear in Analytics. See Balance & billing.
What runs for real in a test conversation
The Test Dialog, the test chat an assistant or Claude opens for you, and every eval run share one rule: reads run live, writes are simulated.
| In a test conversation… | What happens |
|---|---|
| Knowledge, product and web searches, CRM and calendar lookups | Run for real against live data, the same as for a customer. |
| Lead actions (pipeline status, labels), operator handoff and notifications | Simulated — your pipeline stays untouched and no operator gets pinged. |
| Tools that write into a connected system — a calendar booking, a CRM deal, an order, a payment link | Simulated. The tool answers "Simulated in test mode: … nothing was created or changed" and the conversation carries on as if it had worked, so you can rehearse the whole flow without touching the live system. |
| Tools from a connected MCP server that the server marks as read-only | Run for real. |
| Any other tool from a connected MCP server | Not executed. The assistant is told the source is unavailable in test mode and must say so rather than guess — which is exactly what you want to see it do. |
MCP servers declare which of their tools only read (the standard readOnlyHint); the platform records that when you connect the server and again whenever you open its tool list. A server that declares nothing has all of its tools held back in test conversations. To let such a server's reads run, ask its maintainer to add the hint, then reopen the server's tool list under Integrations so the platform picks it up.
Check the tool, not just the answer
A simulated write still says so in the transcript. If a test conversation reports a booking or a deal, look for the "Simulated in test mode" line in the tool result before assuming the integration itself works — for that, use the Test → Run button on the tool (above), which does execute for real.
Model settings (account-level)
There is no per-assistant model or temperature knob. The engine is chosen per account: Settings → AI Core Settings (app.mychatbot.app/profile) — Choose Model from the catalog (priced per 100 AI replies), plus a Reasoning effort control (Low / Medium / High) for GPT-5 models: "higher effort means deeper reasoning and slightly slower replies." Per-assistant tone is what Communication style (above) is for.
Use cases
- Split sales and support: Create copy of your main assistant, rename it, give the copy the warm, empathetic style plus a support-oriented Role, and point each channel at the right assistant on the Channels page — see which assistant answers which channel.
- Collect qualification data hands-free: enable Contact Details + Client Context with fields like Budget Range and Deadline — the assistant fills the client card while it chats, on every channel.
- Night shift: set Work Hours with your timezone, and add Extra working hours for a weekend promo window.
- Human takeover without the customer noticing: enable Implicit operator handoff + Silent handoff, and set a Custom return interval of 120 minutes so the AI resumes if the operator walks away.
- Trim token costs on a big catalog: enable Strip knowledge base results from history and move long policy texts into Dynamic skills.
- Sold out, closing early, running late: add the owner's and the shift manager's phones as control numbers. A voice note such as "The burger is sold out today" applies to every customer chat from the next reply and ends by itself at midnight; "Which notes are active?" shows what customers are being told.
Test it
- Open app.mychatbot.app/configuration, pick your assistant, and use the Test Dialog to ask something a real customer would. Clear chat between scenarios.
- Editing the script? Open the Editor, then the Context panel — type the customer's question and confirm the right knowledge snippets come back before you blame the prompt.
- Verify a tool works with the Tools panel → Test → Run (e.g. run a product search with the exact words a client used) — remember write-type tools execute for real, so use dummy data you can clean up.
- Check handoff: reply to a live chat from Active chats and confirm the chat flips to operator mode — and returns after your interval. To check the stop word or implicit mode, send your stop word (or any message, in implicit mode) from the channel's own app instead.
- Finally, message the assistant through a real connected channel — channel representation settings (link stripping, paragraph splitting) only show their effect there, not in the Test Dialog.
- Control numbers: from the control phone, send "Which notes are active?" to the assistant's WhatsApp number and expect a short answer, not a sales reply. Send an instruction, check it appears under Live notes, then ask about it in the Test Dialog and confirm the reply follows the note without quoting it.
See also
- Sales Platform overview — the map of everything around the assistant.
- How search works — what happens when the assistant queries your product knowledge.
- Follow-ups — building the sequences you attach in the Toolkit.
- Labels & statuses — the pipeline vocabulary Lead Management manages.
- Notifications — everything the Flight Control Telegram group receives.
- WhatsApp — the channels control numbers write to.
- Channels — connecting the places customers actually talk to the assistant; this is also where each channel is assigned to an assistant (and where per-channel welcome messages live).
- Agents Platform — the other kind of agent; don't confuse the two.
