Learn Mampu AIstep by step
A complete walkthrough, from opening an account to curating what the AI knows. Follow it in order and you will have a published AI agent answering on WhatsApp by the end.
60 min in totalFree, no sign-up to watch
Get set up
Open an account, decide what your agent is for, and connect the first channel.
Platform tour, home dashboard and support
Create an account, pick your industry and channels, then walk the left navigation: Conversations, Flow Agents, Contacts, Reports, Blasting, Help Center and Settings. Ends on where to search and where the guides live.
- Create a workspace and finish first-run setup
- Find your way around the main navigation
- Search conversations and contacts
Plan your setup: objectives, channels and use case
The planning step before you build anything. Works through a customer journey, choosing one channel to start with, picking a single use case, then writing an objective that names the audience, purpose, outcome and the details to collect.
- Write an objective your agent can be measured against
- Choose which channel and use case to start with
- Decide when a conversation should go to a human
Connect your first channel and inbox
Settings, Inboxes, Add inbox. Names the inbox, connects a WhatsApp Business number by QR code, sets who has access, then sends a test message to confirm it lands in Conversations.
- Connect a WhatsApp Business number
- Control who can see an inbox
- Confirm the connection with a test message
Build the agent
Draft a flow with AI, teach it your business, then test and publish it.
Create your first AI agent and conversation flow
Describes the journey in plain words and lets the AI builder draft the flow, then walks the block types: send message, condition, collect, and end or handoff. Edits the welcome message, checks the service-routing branches, and tests the result in the playground.
- Draft a flow from a description instead of building it block by block
- Read and edit send, condition, collect and handoff blocks
- Route by what the customer asked for, and hand off on request
AI Knowledge: add, manage and improve answers
Connects an OpenAI key, then fills the knowledge hub: FAQs, products, promotions, scheduling, tone and topics to avoid. Shows why "price / depends" teaches the agent nothing and rewrites it properly, imports an existing document, and tests retrieval with a question worded differently from the FAQ.
- Write a knowledge entry the agent can actually use
- Import FAQs you already have, and review what lands in pending
- Set tone, and tell the agent what not to answer
Test, connect and publish your agent
Runs the agent in the playground against real questions, checks the handoff, then reviews test history to compare runs. Assigns the inbox, sets how long the AI pauses after a handoff, restricts replies to approved testers, and switches the audience to public. Ends on the usage dashboard, where replies, cost per reply and remaining AI allowance are tracked.
- Test a finished flow, and review test history across runs
- Attach an agent to an inbox, and pause the AI when a human replies
- Publish to approved testers first, then to everyone
- Watch replies, cost per reply and remaining AI allowance
Run the inbox
Handle, organise and keep track of the people who message you.
Conversations: replies, private notes and folders
Takes one conversation from first message to resolved. Covers the folders (all, mentions, participating, unattended, mine, unassigned), replying versus private notes your customer never sees, canned responses, and resolve, reopen and snooze.
- Tell replies apart from internal notes
- Use canned responses instead of retyping
- Resolve, snooze and reopen a conversation
Organise conversations: assignment, labels and priority
The actions panel: assign an agent and a team so there is a clear owner, set priority by how fast it needs attention, and label what the conversation is about.
- Give every conversation an owner and a team
- Set priority and labels
Contacts: create, import, search and segment
Adds a contact by hand, then internal notes and custom attributes for the things your business tracks. Imports a CSV, searches by name, filters by field, and saves the filter so the team can return to the same group without rebuilding it.
- Store what matters with custom attributes
- Import an existing database from CSV
- Save a filter as a reusable segment
Reach out and book
Send campaigns, set up multi-step follow-ups, and take appointments.
Blasting and message sequences
Builds the audience with a label first, then a campaign with media, a live audience count, a schedule and batch settings. Then the difference that matters: a blast is one message to many, a sequence is several follow-ups over time, with delays and conditions on each step.
- Build a campaign audience from labels
- Schedule a blast and send it in batches
- Choose between a blast and a multi-step sequence
Channel blasting: WhatsApp and live chat
The same campaign on two channels. WhatsApp Business: message, inbox, audience label, schedule, batch and delay. Website live chat: sender, the URL it applies to, and a time-on-page condition before it shows.
- Schedule a WhatsApp Business campaign
- Trigger a live chat message by page and time on page
Calendar booking and scheduling
Sets up a calendar: members, business hours, appointment length, then the settings that decide what customers can actually book, slot interval, buffer, minimum notice and booking horizon. Tests the public booking link as a customer, then verifies the appointment and blocks out lunch.
- Set business hours, buffers and minimum notice
- Share a booking link and test it as a customer
- Verify, cancel and block out time
Measure and improve
Read the reports, and turn what they say into a change.
Reports: conversations, agents, teams, CSAT and SLA
The manager's guide. Live overview of open, unattended, unassigned and pending, then historical volume, first response and resolution times, comparisons by agent, inbox, team and label, and what CSAT and SLA figures are telling you. Closes on a worked example: unattended peaks between noon and 2pm, so move coverage and check again next week.
- Read first response and resolution time properly
- Compare agents, inboxes, teams and labels
- Turn a CSAT or SLA number into a staffing decision
AI agent reports and exports
Bot performance over a date range, flow submissions for completed conversations, choosing which fields the submission table shows, and downloading the data for analysis outside the platform.
- Review AI agent activity over any period
- Inspect flow submissions and pick the columns
- Export report data
Connect more channels
Take the agent past the first number: the three WhatsApp setup paths, then TikTok, Telegram and Facebook.
WhatsApp setup paths, migration and account health
Which of the three connections you need, and why: WhatsApp Business links an existing Business app by QR code, WhatsApp Cloud runs through Meta's own onboarding, and migration is for a number already on the API with another provider (two-step verification has to be off first). Ends on Meta-side billing, getting a restricted number reviewed, and permanent deletion.
- Pick between WhatsApp Business, Cloud and migration
- Warm a new number up instead of blasting it into a restriction
- Request a review if the number does get restricted
TikTok: connect, migrate and auto-reply to comments
Connects a TikTok business account, and shows the step people miss when moving one across: revoke the old platform first under TikTok's apps and services. Then comment-reply rules, where video keywords decide which videos a rule covers and trigger keywords decide which comments fire it.
- Connect a TikTok business account, or migrate one over
- Limit a rule to videos whose caption matches a keyword
- Reply publicly under a comment and like it automatically
Telegram and Facebook, with comment automation
Creates a Telegram bot in BotFather and connects it with its token, then adds a Facebook page. Builds a comment rule on the keyword "appointment" that answers publicly and opens a private Messenger thread, attaches the Facebook inbox to a flow agent, publishes it, and tests the whole thing with a real comment.
- Create a Telegram bot and connect it with its token
- Answer a comment in public and continue in private
- Attach an inbox to a flow agent and publish it
Run the workspace
Publish a help centre customers can search themselves, manage the account behind it, and set up the agents, teams and reply targets.
Build and manage your help centre
A portal end to end. Categories with slugs and descriptions, an article built with headings, lists, images and links, held as a draft so the team can review it, previewed exactly as a customer sees it, then published and checked on the live portal. Closes on a second language added as a draft, plus brand colour, custom domain and layout.
- Organise articles into categories customers can navigate
- Draft, review and preview before anything goes public
- Add another language, a brand colour and a custom domain
Account settings, monthly active contacts and referrals
Account name, site language, and the account ID that API integrations need. Then billing, where monthly active contacts are counted separately from your total contacts and tracked against the plan limit, and the referral dashboard with its shareable link.
- Change the account name and language, and find the account ID
- Watch monthly active contacts against the plan limit
- Copy your referral link
Agents, teams, automation, macros and reply SLA
Invites agents with a role and groups them into teams so conversations reach the right one, then the reusable pieces: labels, custom attributes, canned responses, and macros — which unlike a canned response carry out several actions at once. Builds a rule that assigns, labels and acknowledges a new conversation on its own, sets the reply targets, and reads the resulting SLA report.
- Invite agents and group them into teams
- Build a rule that assigns, labels and replies by itself
- Set first response, next response and resolution targets
Connect your tools
Push data out to the systems you already use, take it in from them, and bring your own AI keys.
Google Sheets and inbound webhooks
Connects a Google account and picks the spreadsheet that completed flows write into, where a returning contact updates their existing row instead of adding a second one. Then inbound webhooks: pick the provider and target inbox, map the incoming payload onto contact fields, choose what the webhook does, and fire a test request at the endpoint before trusting it.
- Send completed flow submissions straight into a spreadsheet
- Map an incoming payload onto contact fields
- Test a webhook endpoint before relying on it
Connect OpenAI, Claude and DeepSeek keys
Which provider does what: Anthropic and DeepSeek run flow agent replies, while OpenAI handles FAQ and knowledge generation, transcription, image understanding and article translation. Then the same careful routine for each — work in the right project, create a dedicated key, check the account has credit, save, confirm connected, run a small reply test — and what to check when one will not connect.
- Know which provider powers which AI function
- Create a dedicated key and confirm the account has credit
- Revoke and replace a key that has been exposed
Work with the AI coding agent
Give an AI coding agent scoped access to the workspace, then put it to work on knowledge and campaigns.
Connect an AI coding agent over MCP
Creates an access key scoped to read conversations only, with an expiry so a test key cannot outlive its purpose, pairs it with the account ID that says which workspace a request belongs to, and connects Claude Code to the MCP endpoint. Recommends a read-only request as the first thing to try, then covers what further scopes unlock, how each key's last use and call count are tracked, and where the agent is deliberately fenced off.
- Create a key scoped to only what the task needs
- Connect Claude Code and verify with a read-only task
- Know the limits: a draft still needs a person to send it
Knowledge curation, AI-assisted blasting and troubleshooting
Mines resolved conversations for the questions customers actually asked, turns them into FAQ drafts, and holds the line on reviewing them — approve only what you want the AI to use, then test it in the playground with a realistically worded question rather than assuming it works. Then a WhatsApp blast the agent drafts but a person sends, and a troubleshooting pass over permissions, authorization, quota and template status.
- Turn resolved conversations into reviewed FAQ drafts
- Let the agent draft a blast, keep the send with a person
- Diagnose permission, quota, authorization and template failures
