AI Agents #

An AI Agent is the brain behind your bot — its personality, what it knows, its primary language, and how it behaves in a conversation. Where a campaign used to bundle “who the bot is” together with “who it messages and when,” an Agent holds just the “who the bot is” part — so you build it once and reuse it anywhere. Campaigns have retired; an Agent plus a Broadcast is what replaced them.

Think of an Agent as the team member you’re training: give it a name, a personality, product knowledge, and a set of tools, and it can then be put to work handling replies on any channel, and on any Broadcast that points to it.

Where Your Old Campaign Bots Went #

Every campaign’s bot is now an Agent — nothing was lost and you didn’t have to do anything. If you had campaigns with a configured bot, each of them has a matching Agent holding the exact same instructions, FAQs, and settings, and that Agent is where the bot is edited from now on.

The Agent is the single source of truth. Change its instructions and the change reaches every conversation it answers, on every channel, immediately — there is nothing to copy into a second place.

Imagine this: You update your Agent’s instructions to mention a new return policy. Every conversation that Agent handles picks it up straight away.

Old campaign links still work: opening one lands you on that campaign’s Agent editor. See Campaigns Have Moved to Broadcasts & Agents.

Where to Find AI Agents #

In the main menu down the left side of the screen, under AI Studio, click AI Agents. (On a phone the menu is hidden — tap the menu icon ☰ in the top corner first, then tap AI Agents.)

The AI Agents page lists every Agent on your account, including one automatically named after each of your old campaigns’ bots, along with a Who answers new conversations panel above the list — a one-line summary, or a row per connected channel, showing which Agent picks up a brand-new contact on that channel. See Entry Points → Who answers new conversations for how that panel works and what it means for your live routing today.

What Lives on an Agent #

An Agent groups together everything that shapes how the bot thinks and talks:

An Agent does not hold a contact list, an opening message, a channel, or a send schedule — that’s what a Broadcast is for. Appointment booking is set on the Agent, on its AI Abilities tab — see Booking System.

Creating an Agent #

Click New agent at the top of the AI Agents page and pick a starting point — the dialog offers up to three:

1. Build from your website (the default, and the fastest way to a working agent): type your website address and we write the agent for you. Click Find pages to see a list of the pages on your site and tick the useful ones — a handful of relevant pages is pre-selected for you — or skip that and we’ll read the address you typed. You can also add documents (price lists, brochures, FAQs) for the agent to learn from. Give it a name — one is suggested from your website — check the language, and click Build agent.

   The dialog then stays on screen while your agent is built: it shows what’s happening (reading your website, then writing the instructions) and usually takes a minute or two. When it finishes, you land in the agent’s editor with the instructions and knowledge already filled in — review them, tweak what you like, and you’re done. Don’t feel like waiting? Click Continue in the background and carry on with other work — the agent shows a Building… badge in your list until it’s ready, and opens once the writing has finished.

   Building the agent takes a minute or two, but importing every page you ticked can take longer. When you tick more than 25 pages or files, the dialog says so with a rough time. You can open and test the agent before that import is done: it answers from the pages that are already in, and the rest is added as each page finishes.

   If a build can’t finish, the dialog tells you why in plain words rather than leaving you watching a spinner — the website couldn’t be read, there wasn’t enough text on the pages to write an agent from, or the account has run out of credits. Fix the thing it names and start the build again.

2. Start blank: give the Agent a name and pick its primary language, then click Create agent. You’ll land straight in its editor to fill in instructions and knowledge yourself.

No incoming/outgoing choice here — that’s not a missing step. Classic campaigns made you pick Incoming, Outgoing or Combined up front. An Agent doesn’t, because that split moved off the Agent itself: an Agent is the side that answers people who message you, and reaching out first is what a Broadcast does. So the old “Combined” setup is now an Agent plus a Broadcast that picks it as the reply handler — you build the Agent here, then create the Broadcast when you’re ready to send.

English or English (UK). Both are in the language list. English writes American spelling. English (UK) writes British spelling and vocabulary (colour, organise, mobile, flat), for businesses whose customers should not hear an American voice. Like any primary language, it does not lock the agent: a contact who writes in another language still gets an answer in their own language. Anything you write about how the agent should speak (tone, words to use or avoid, a regional style) goes in Persona & Goal or Conversation Behavior on the AI Instructions tab, and the agent follows it on top of the language you picked.

Mixed languages written in Latin letters. For contacts who text their own language in English letters mixed with English, pick the matching option: Singlish (Sinhala + English), Hinglish (Hindi + English), Taglish (Filipino + English), Roman Urdu (Urdu + English), Banglish (Bengali + English), Tanglish (Tamil + English), Manglish (Malay + English) or Arabizi (Arabic in Latin letters). The agent then answers the way these contacts write, for example “June 26 wala room rates oya select karana room category eka anuwa wenas wenawa”, always in Latin letters and never in the native script, and it follows each contact’s own mix. A contact who writes plain English still gets plain English. For every option except Taglish, chat summaries, alerts and generated FAQs are written in English.

Filipino, Taglish and Cebuano. All three are in the language list. Filipino writes proper Filipino; Taglish (Filipino + English) writes the everyday Tagalog-English mix people actually type in chat. If your customers message you in Taglish, pick Taglish — an agent set to Filipino answers correctly but in a more formal register than the conversation is in. Cebuano (Bisaya) is for customers in Cebu, the rest of the Visayas and much of Mindanao who write in Bisaya rather than Tagalog — pick it and the agent answers in everyday Cebuano instead of drifting into Filipino.

Looking for “Extract from a campaign”? It’s gone — there are no classic campaigns left to extract from. Every campaign bot became an Agent automatically.

Editing an Agent #

Open any Agent from the AI Agents list to reach its editor. The steps run down the left, and each one is a short stop — here’s the tour in order. An agent that is already active opens on its Overview (its numbers); a paused or brand-new agent opens on AI Instructions, because that is where its work is.

1. Overview — the agent’s own dashboard: AI replies, replies and reply rate, new contacts, appointments booked and credits used for the last 7, 30 or 90 days, each with a small trend line and, where there is a baseline, the change against the period before; its full readiness scorecard with the fix for every gap; the channels it answers on, its median first-response time and the share of its messages answered by AI rather than your team; then the same Conversation outcomes, AI vs human and Activity panels the Dashboard shows, limited to this agent. See Agent overview.

2. AI Instructions — the primary language, persona, playbook, response pacing, message limits, and model (AI Speed/AI Quality) settings, all on one tab. This is where most of your agent’s personality lives — see AI Bot Setup for every field.

Test mode is the first panel on this tab: switch it on and the AI only answers the contacts on its test list, so you can try the agent on a real channel safely. See Test mode.

Primary language sits at the top of this tab. Change it and it saves straight away — you don’t need to publish it like the rest of the tab, and it takes effect on your live conversations immediately. It changes the language your agent writes in from here on; anything already written (its instructions, its FAQs) stays in the language it was written in, so retranslate those yourself if you want the whole agent to switch.

3. Active Hours — when the agent is allowed to reply. It starts as Active 24/7; switch that off to set a weekly schedule. Outside the hours you pick the agent stays quiet, and anything that comes in is answered automatically the moment the next window opens — those chats wait under the Outside hours filter in Chats until then. See Bot Availability Schedule.

4. Media — the pictures, videos, voice notes and PDF documents the agent can send mid-conversation. The tab opens on this agent’s own media and works the same on a brand-new agent as on an established one: upload a file here and it belongs to this agent straight away, with nothing to attach first. See Media Library.

5. FAQs & Knowledge — what the agent knows: its FAQs, linked knowledge sources (websites and files), live web search, and any extra context. This is the single biggest lever for answer quality — see Managing FAQs.

Once an agent is linked to a few hundred pages, use the search box above the Knowledge sources list to find one: it looks at the website, the folders inside it and the individual pages all at once, so typing part of an address (alternatives, say, or pricing) narrows the whole tree to the matches and opens the folders they sit in. Each website then shows how many of its pages matched. One thing worth knowing: while a search is active, a website’s or folder’s Unlink all button becomes Unlink matching and only unlinks the pages you can actually see — the ones filtered out by your search are never touched.

To unlink several pages that don’t share a folder, use the checkboxes: every page row has one, and each website and folder header has a select-all that ticks everything inside it (a dash in the box means only some of it is ticked). As soon as anything is ticked, a bar appears at the bottom of the screen counting your selection, with a single Unlink button — you confirm once and the whole selection is detached together. The per-page remove button and Unlink all still work exactly as before.

6. AI Abilities — what the agent is allowed to do: book appointments, call your own systems through Custom Functions, use MCP server tools, update contact details, and create tasks. The Bookings panel here is where you let the agent schedule meetings and pick which event types it can book — one or several, so an agent can handle every service you offer — see Booking System.

7. Tags — let the agent label contacts as it chats (for example “Interested”, “Booked”, “Not a fit”). Describe when each tag should be applied and the agent handles the rest — you can filter and segment contacts by these tags later. See Creating Tags. Rules only run while the agent is replying, so conversations that already happened — or that your team handled with the AI paused — don’t get tagged on their own: click Apply to existing conversations on this tab to run the rules over them. See Tagging a conversation on demand.

Don’t want to write the tag rules yourself? Click Generate with AI on the Tags tab: the platform reads this agent’s instructions and business context and proposes a set of tags (typically three to eight) with a ready-made “when to apply” rule for each — lifecycle stages like Interested or Not a fit, facts the conversation reveals, and hand-off reasons. It checks your existing tag library first and reuses a tag when one already means the same thing, so you don’t end up with duplicates. Pick Add to my tags to keep what’s already there and only add new ones, or Replace all tags to start over. A run costs 1 credit (0.25 on the Max tier, free with your own AI key), and rows the AI wrote carry a small AI marker until you edit them. Agents created from a website or documents — in the setup wizard or with New agent — get their tags generated automatically as part of the build, so this tab is usually already filled in when you first open it.

8. Follow-ups — how this agent chases contacts who go quiet: quick nudges within the first 24 hours, warm-lead cycles for contacts who engaged and then went cold, and cold-lead messages for contacts who never replied at all. Off by default. Pick the WhatsApp template for each follow-up slot from your template library. See Automated Follow-Ups for what each type does and how to time them.

Don’t want to write those messages yourself? Turn follow-ups on and click Generate AI templates: the platform writes a set of follow-up messages from this agent’s own instructions and business context, shows you the progress as it goes, and submits them to WhatsApp for approval. It works on any agent — one you created a minute ago, a copy, or one that arrived in the account already built. See Auto-Generated Templates.

9. Entry Points — routing rules for this Agent; see Entry Points for what they do and when they go live.

10. Try Out — a sandbox chat where you play the customer and test the agent’s replies before real contacts ever see them. It runs on this agent directly, so there’s nothing to attach before you can test, and each agent keeps its own test conversation. See Chat Playground.

Agent overview #

Every Agent has an Overview step — the first step in its editor, and where an active agent opens when you click it in the AI Agents list. It is the per-agent version of the Dashboard, the same way a Broadcast has its own stats page.

• The numbers. Pick Last 7 days, Last 30 days or Last 90 days at the top. The cards show AI replies, replies (with the reply rate underneath), new contacts, appointments booked (only when this agent has booked any in the range) and credits used. Each card carries a small trend line and, where there is a previous period to compare against, the change since it. Only this agent’s activity counts — a contact moved to another agent later keeps its history where it happened. What each card counts: AI replies is every message the AI sent through this agent, counted per bubble, so an answer the AI splits into three bubbles adds three. Replies is the messages contacts sent back. New contacts is contacts created through this agent’s conversations. Appointments booked is the bookings it made. Credits used is what its replies and tool calls spent. AI replies is normally several times Replies, because one inbound message usually gets a multi-bubble answer and follow-ups count too. Hover the small i on any card for its exact definition.
• Readiness. The full scorecard: the score ring, the traffic light, the six section bars, and every check under Needs attention or What is working, each with the button that fixes it. See Readiness score for what is scored.
• Where it answers. The channels its enabled entry points route to it, its median first-response time over the range, and the share of its messages that the AI answered rather than your team. Entry points opens the routing step.
• Conversation outcomes, AI vs human, Activity. The same three panels as on the Dashboard, limited to this agent: conversations resolved without a human, escalations, unhappy customers, the questions it couldn’t answer and the sales left open, the AI-versus-human split, and the latest replies, bookings and alerts.

Nothing on this step uses credits; it is all plain counts over data your account already stores. A block whose source is unavailable says so rather than showing zeros.

Readiness score #

Every Agent carries a readiness score from 0 to 100 — a quick answer to “is this agent set up well enough to talk to real customers?” without opening every step and checking by hand. You see it in three places:

• The AI Agents list has a Readiness column showing each agent’s score with a coloured dot. Click a score to jump straight to that agent’s scorecard. Above the list, the Any readiness / Ready / Needs attention / Action required chips filter the list by colour (with a count on each), and the issue picker next to them narrows it to one specific gap — agents with no entry point, no FAQs, paused, never tested, booking not wired up, follow-ups incomplete, a channel that isn’t connected, or missing instructions. Agencies managing many client agents can find the ones that need work in seconds.
• The agent editor header shows the score next to the Active/Paused pill. Click it to open the Overview, where the scorecard lives.
• The Overview step (the first step in the editor) carries the full scorecard under the numbers.

What the colours mean #

• Green — Ready (80 and above, nothing failing): the agent is configured, connected and answering.
• Amber — Needs attention (50 to 79, or a check that fails): the agent works, but something is missing or half-done.
• Red — Action required (below 50, or a critical check failing): the agent cannot do its job. A critical check is one that stops the agent outright — no entry point, paused while conversations are still routed to it, bookings switched on with booking language but nothing to book into, a routed channel that isn’t connected, or no instructions at all.

What is scored #

The score is made up of six sections, each worth a fixed number of points:

Each check shows how many points it earned, a one-line explanation, and — when something is wrong — a button that takes you to the step where you fix it (or activates the agent on the spot). The scorecard also names the single most valuable thing to do next. Click a score in the AI Agents list or the badge in the editor header and you land on it. Checks that don’t apply to your agent (bookings not used, no external tools attached) are shown for completeness and don’t cost points.

A few things worth knowing:

• The score looks at what is live. Draft changes you haven’t published yet don’t count until you click Publish; the scorecard says so, and Recheck re-runs it after you publish.
• An agent used only through its website embed code has no entry point row, so the routing check flags it. The embed routes on its own, so you can ignore that one check for such an agent.
• An agent that should not book but has the booking switch on (a leftover from an older setup) shows “Bookings are on but not set up” with a Turn bookings off button. Click it and the check clears; the agent was never booking anything in that state anyway.
• Some checks are educated guesses, and you can overrule them. “Booking is mentioned but not enabled” looks for booking instructions in the text — “book a call”, “offer them a slot”, “plan een afspraak” — and ignores booking words on their own, descriptions of what the product does (“the AI books appointments”) and negated rules (“never offer a call”); it can still misread an unusual phrasing. The same goes for “asks for details it cannot save” and “pricing not mentioned”. Those three carry a Not relevant for this agent button: click it and the check counts as passed from then on, shows an Ignored tag under What is working, and has a Check again link if you change your mind. None of these three can turn an agent red on their own.
• A red score does not switch anything off. It’s a report, not a gate — the agent keeps answering exactly as it did before.
• Over the API, the same report is available as GET /v1/agents/health (every agent) and GET /v1/agents/{agentId}/health (one agent), so an agency dashboard or an assistant can read it too.

Looking for the Attached Campaigns step? It’s gone. It was a read-only list of the campaigns an Agent answered for, and there was never anything to set on it — what decides which conversations reach an Agent is its Entry Points, and a Broadcast chooses its own reply handler when you build it. Publishing a change to the Agent still reaches every conversation it answers, immediately, exactly as before.

Changes to how the agent talks, meaning its instructions and its reply settings (AI speed, typing speed, response delays, max messages and the like), are saved as a draft: the Try Out chat uses it straight away, but real conversations keep running on your previous setup. While a draft exists, an amber banner at the top of the editor shows what’s unpublished — click Publish to live when you’re happy with how the agent replies, or Discard to throw the draft away and go back to the live setup. This means you can safely tweak and test a live agent’s wording without customers ever seeing a half-finished change.

Everything else saves instantly and is live for your customers the moment you save it: the agent’s name, switching it on or off, its language, test mode and its contact list, linked FAQs, knowledge sources, knowledge groups and AI web search, custom functions, MCP servers, bookings and Odoo, follow-ups, tags, the media library and the Instagram Story and share reply settings. These tabs and panels carry a small Saves instantly tag next to their heading, so you can always tell which kind of change you are making. The banner says the same: drafts cover instructions and reply settings, and the panels marked Saves instantly go live right away.

To see exactly what the draft changes, click View changes in the banner. Each changed setting is listed with the live wording against the draft: removed text in red and struck through, added text in green, and switches as On or Off. Publish to live opens the same list, with a Show changes toggle on each setting, so you can check the before and after one last time before you confirm.

Optimize with AI sits in the Try Out chat panel’s header and is available on every agent — a copy you just made, one that arrived in the account already built, or one you’ve been running for a year. Describe what you want improved (or click the thumbs-down under a reply that missed the mark) and the agent’s instructions are rewritten for you, with a side-by-side review of every proposed change before anything is kept. What you accept is saved as a draft like any other edit, so it reaches real conversations when you click Publish to live — not a moment before. See Chat Playground.

Version history and undoing a change #

Every time you click Publish to live, DM Champ saves a version: which settings changed, how they read before, and how they read after. An Optimize with AI run is recorded too, with the changes it proposed. Ordinary saves are not, because a save only touches your draft, so the list stays a short record of the changes that mattered.

Open it with the Version history button at the top right of the agent editor, next to Optimize with AI. The button is there whether or not you have unpublished changes waiting. Each entry shows whether it came from a publish or from an AI optimisation, when it happened, and which parts of the agent it touched. Click Show changes on an entry to see what changed, field by field, in plain words. Long text such as instructions or the persona shows the old and the new wording line by line. Switches read as On or Off, and an empty setting reads Not set. For a group of settings, such as the follow-up settings, only the settings that actually changed are listed, each with its old and new value (for example Cold lead wait times: 72h → 72h, 168h). Click Show raw on such a group to see exactly what was stored, which can help when you are talking to support. The newest 20 entries load first; click Load older versions at the bottom of the list to see the ones before them.

To go back to an earlier setup, click Restore as draft on the entry you want. The values from before that change are loaded into your draft, and nothing goes live yet. From there it is the normal flow: test the agent in Try Out, then click Publish to live when you are happy, or Discard to throw the restored draft away and keep what is live today.

Settings marked Saves instantly never go through the draft. When one of those changes, the list gets an entry badged Changed live so you can see what changed and when. These entries have no Restore as draft button, because there is no draft to restore; change the setting back in the editor instead.

Changes the in-app helper makes to an agent for you land in the draft as well, so they show up in the amber banner and wait for you to publish them, exactly like an edit you typed yourself.

Turning an Agent On or Off #

Every Agent shows as Active or Paused — on the AI Agents list (see the screenshot above) it’s a pause/play icon button in the row’s actions, and on the Agent’s own editor it’s a status badge next to the name. It’s a hard kill switch: when an Agent is turned off, it stops replying everywhere it’s used — this cannot be overridden by a per-contact “AI on” setting. It applies everywhere that Agent is used: every channel it answers, and every Broadcast pointing at it.

Test Mode: Let the AI Answer Only Your Test Contacts #

Test mode lets you try an agent on a real WhatsApp number, Instagram account or any other channel without any risk to your real customers. While it is on, the agent stays connected to its channels, but the AI only answers the contacts you put on its test list. Everyone else is left completely alone by the AI.

What still happens for contacts who are not on the list:

• Their messages arrive in Chats as usual, so nothing is lost.
• You and your team can reply to them yourselves.
• The AI does not reply, does not send them follow ups, and does not answer their Instagram or Facebook comments.

Broadcasts, WhatsApp templates and messages you send yourself are not affected by test mode.

Turning test mode on #

1. Open the agent from the AI Agents list.
2. Go to the AI Instructions step. Test mode is the first panel at the top.
3. Switch Test mode on.
4. Under Contacts the AI may answer, pick the channel, type the phone number, email or handle and click Add. You can paste several at once, separated by commas.

Every change saves straight away, and there is nothing to publish. Each change also adds a Changed live entry to the agent’s Version history, so you can see when test mode was switched or the list changed. The agent then shows a blue Test mode badge next to its status on the AI Agents list and on its editor, so nobody forgets it is on.

If you switch test mode on with an empty list, the AI will not answer anyone until you add a contact.

What to put on the list #

Pick the channel first, then add whatever identifies the contact on that channel:

• Phone (WhatsApp, SMS): the phone number with its country code, for example +31 6 1234 5678. Spaces, dashes and brackets do not matter, and 0031... works too. A number without its country code does not match. A phone number counts on WhatsApp and SMS only. A WhatsApp contact that hides its number can be added under Phone by its WhatsApp user id instead (two letters, a dot and a long code, for example US.13491208655302741918).
• Instagram, Messenger, Telegram, LinkedIn and the other channels: the handle (with or without the @), the email or the account id you see on the contact.

Each entry only counts on the channel you picked for it, because @john on Instagram and @john on Telegram are two different people. To test the same person on two channels, add them once per channel. The list shows the channel next to each entry.

Entries added before channels existed show Any channel. They keep working on every channel except WhatsApp and SMS; remove one and add it again with a channel to limit it.

A contact matches when its phone number, email, handle or id is on the list for the channel it wrote in on. The list holds up to 200 entries.

Going live #

When you are happy with the agent, switch Test mode off. From that moment the AI answers everyone again, exactly as before. Your test list is kept, so you can switch test mode back on later without typing it again.

Renaming, Duplicating and Deleting an Agent #

• Rename an Agent by clicking its name at the top of the Agent editor — type the new name and press Enter. The rename applies immediately, everywhere the Agent is shown.
• Duplicate an Agent with the copy icon (tooltip Duplicate agent) on its row in the AI Agents list to start a new one from a copy of its current settings. The copy keeps the original’s on or off state, so a copy of a live Agent is live too. Files in the original’s Media Library are not copied: upload the ones the copy should send. The copy opens in the editor named “(copy)” — click the name to give it its own. The copy starts with an empty Try Out conversation of its own, so testing the copy never appears in the original’s test chat (or the other way round).
• Delete is blocked while an Agent is still used by a Broadcast or an Entry Point — point those at another Agent (or remove them) first, then delete. An Agent nothing is using deletes right away. An Agent that came over from an old campaign together with a Broadcast of the same name is a special case: that Broadcast is a view of the same campaign, so it does not block the delete, and you can also delete the Broadcast itself from the Broadcasts page (unless it is currently sending or scheduled). Watch for Entry Points you didn’t add by hand: picking the comment to DM use case in the setup wizard creates an Instagram and a Facebook comment rule on the new Agent, and those count as “in use” until you remove them on the Agent’s Entry points tab.

Switching an Agent On or Off From the List #

The play and pause icon in a row’s actions switches that agent on or off without opening it. Pausing happens straight away. Switching a paused agent on asks you to confirm first, because it starts answering real contacts on its channels at once; on a WhatsApp Web line that is your own number. Test it in Try Out before you go live.

Working With Several Agents at Once #

Each row on the Agents page has a checkbox (the header checkbox selects everything currently shown). With agents selected, a bar appears at the bottom with Activate, Pause and Delete — each asks you to confirm before anything happens. Deleting skips any agent that’s still in use by a Broadcast or an Entry Point; free those up and the next delete takes them too. Press Escape to clear a selection.

Next Steps #

• Broadcasts — send a one-off or ongoing message and choose which Agent handles the replies.
• Entry Points — routing rules that decide which Agent picks up an incoming conversation, and whether they’re already live for your account.
• Setting Up Your AI Bot — the full guide to writing instructions, playbooks, and response settings.
• Managing FAQs — give your Agent ready-made answers.