Guides
How it actually works, written down.
These are operator guides: what a channel really enforces, why an automation stopped firing, and what to do about it. Where what we measured differs from what the platform documents, both are shown and labelled.
Start where your problem is.
Comment automation, the one-DM budget, and the rules that actually bind.
What Instagram requires before you can automate comments Account type, the one setting everyone misses, and what you do not need — in plain terms, before you sign up for anything. Set up your first Instagram comment automation Connect your account, then describe what you want in a sentence. No flow builder, no dragging nodes. Comment trigger options: keywords, matching, and which posts Every option on an Instagram comment trigger, what it does, and which one you want — with worked examples of what matches. The Instagram comment-to-DM rules that actually bind What Meta's comment automation really allows, measured against a live business account — including one documented behaviour that did not hold. Why your Instagram comment automation stopped working The causes, ordered by how often they actually happen — starting with the account setting no tool can fix for you. What happens after the DM Comment automation delivers one message. The lead is in the conversation that follows — and most tools have nothing to hold it with.
Templates, the service window, and keeping the WhatsApp Business App on the same number.
Templates
When a WhatsApp message has to be a template The 24-hour window decides whether you can type a reply or must send an approved template. What the inbox shows when it closes, the three template categories, and why a marketing template can fail to arrive. Get your WhatsApp templates into Omni Octo Create templates in Meta's WhatsApp Manager or by asking Octo, pull them in with Sync from Meta, and read the status each one shows — including what to do when an approved template is missing. Send a template from a conversation Pick an approved template inside a WhatsApp thread in the inbox, fill in its variables, check the preview and send it. Where the picker lives, what it asks for, and what each refusal means. Send a template from your own system One API call sends one approved WhatsApp template to one number — from a webhook, a cron job or a script. The key, the template id, the request, every refusal, and the two things to design around. Ask Octo to send a template Tell Octo in chat who to message and what to say, and it finds the approved template, fills in the variables and prepares the send for one WhatsApp number. Nothing goes out until you confirm. What it checks and what it refuses. Send a template to a list A WhatsApp campaign sends one approved template to everyone on a list, from one number. The seven sections of the composer, what each person's row ends up saying, and how it is paced. How fast one WhatsApp number can send Every WhatsApp number has per-second and per-minute limits (and, where a rule sets one, per-hour), plus a daily template limit. The defaults, how the daily cap maps to Meta's messaging tiers, where to change it, and what each way of sending does when the limit is reached.WhatsApp Coexistence
Keep using the WhatsApp Business App on the same number Connect the number you already run in the WhatsApp Business App, keep answering from your phone, and let your AI agent answer the same number. The connect page step by step, and what changes in the app on your phone. Bring your WhatsApp Business App chats and contacts across The one-time import that follows a WhatsApp Business App connection: what comes across, how to read its status on the channel page, the notice when it finishes, and where the imported chats and contacts end up. When you reply from the phone, the AI steps back On a WhatsApp Business App number, a reply you send from the app pauses the AI in that chat. What counts as a reply, how the inbox shows it, and how the chat comes back to the AI after 15 quiet minutes. When WhatsApp disconnects your Business App number Why Meta disconnects a WhatsApp Business App number from Omni Octo, how you are told, what stops while it is disconnected, and how to reconnect — including what reconnecting does to the one-time import.Mailbox shapes, the authentication that gates them, and why a reply never arrived.
Connecting a mailbox
What you need before an AI can answer your email A mailbox, an agent to answer it, and knowing which of seven shapes your mailbox is. The third one decides whether you need an administrator. Connect Gmail to an AI that reads and answers your email Four screens and one checkbox that breaks everything if you miss it. Works for a personal @gmail.com and for a Google Workspace address. Connect a Google Workspace shared inbox (and what your admin has to approve) A real Workspace user mailbox your team shares, authorised once by a super admin through Domain-Wide Delegation. Not a Google Group — that is a different setup. Put an AI on a support@ or sales@ Google Group — no admin needed A group address is answered by watching one member's mailbox and filtering on the List-Id header. The member has to already be in the group; nothing else needs approving. Send email as an alias: connect a Gmail "Send mail as" address The mailbox you sign into keeps receiving the mail; outbound replies go out as the alias. The alias has to be verified in Gmail before you start. Connect Outlook or Microsoft 365 to an AI email assistant One consent screen, seven permissions, and no administrator. Covers a personal Outlook.com inbox and a single-user Microsoft 365 work address. Connect a Microsoft 365 shared mailbox: Full Access and Send As explained A shared mailbox has no password to sign in with. You connect as yourself, an admin grants two permissions, and the wizard refuses to finish without both. Connect a Microsoft 365 Group so an AI can answer its shared inbox A mail-enabled M365 Group, connected through a member. Two one-time grants in the Exchange admin center, and drafts that live on the dashboard rather than in Outlook. Why your Gmail or Outlook connection isn't working Failures by stage — sign-in, consent, naming the mailbox, verification, and the ones that appear weeks later. Each has its own message, and the message is the diagnosis.Answering mail
Can AI answer my emails? Draft, autonomous and human-only, explained Yes, in one of three ways, set per mailbox and overridable per thread. The same three settings are worded differently in setup and in the inbox, which is worth knowing before you go looking. Let AI draft your email replies and approve them before they send Draft mode writes the reply and stops. Where the draft waits, what it tells you about its own sources, and why it is also sitting in your Drafts folder. Make the AI answer email from your website and your documents Grounding is what turns a plausible reply into a correct one. Add a website, upload files or paste text on the agent's knowledge base, and every draft tells you what it drew on. Summarise a long email thread Ask Octo in chat — there is no Summarise button in the inbox. The summary is written in the background as the thread grows, so the answer comes back without waiting on a model. What an AI email assistant can't do with your inbox The structural limits, not the faults — what this will never do, so you can plan around it instead of discovering it in week three. Why the AI didn't reply to an email Eight reasons, in the order worth checking. Most of them are a rule doing its job and saying so on the message — not a delivery failure.Numbers, phone menus, voicemail, and where the minutes go.
Answering calls
What you need before Octo can take calls A number, an agent, and a binding between them. Where each lives, and why a number that answers is not proof that anything is set up. Set up an AI phone receptionist Describe the job in a sentence, take the test call, then tune what you heard. Where each piece lands in the dashboard, and the question Octo will not guess. The limits that actually bind on calls you receive Almost none of them, and that cuts both ways. Where the caps live, what they do not cover, and the one gap we have not closed. Why your voice agent isn't working Ordered by how often each one is the answer. The first is the one that looks like success, and it accounts for most reports. What happens after the call Where the call lands, what gets pulled out of it, and why a Dialog Insight attached to the agent is worth more than one attached to a campaign.Making calls
What you need before Octo can call anyone A list with the right phone field, a signature, and a timezone. The contact-field distinction that silently removes people from your audience. Run your first call campaign Nine sections of the composer, in the order you meet them, with the field that decides whether the results are usable and the two settings that quietly do nothing. Campaign types, and what each one is for Four cards on the New Campaign screen, and only one of them holds a conversation. Which to pick, and the two features that share the words voicemail drop. The limits that actually bind on a campaign Three concurrency ceilings, a schedule window, per-number pacing and a retry policy. Where each lives, and the four rules that surprise people. Why your campaign isn't dialling It says Running and nothing is happening. Eight causes ordered by frequency, starting with the two that aren't faults, and where each one shows itself. What happens after the campaign A results summary that aggregates what every call found out, an export, and the two outcome groups that mean different things about your list.Your agent's voice
Voice pipelines: what to pick and what it costs Three options on one dropdown, the model family each carries, and the ladder to climb when the agent mishandles a phone menu. ElevenLabs voice models: Turbo v4, Conversational v3, Flash v2.5 and Multilingual v2 The same voice can sound four different ways. What each ElevenLabs model sounds like on a phone call, how fast it starts speaking, which languages it cannot speak, and what it costs. Use your own ElevenLabs voice on your phone agent Save your ElevenLabs key in Integrations, import your cloned or designed voices, and your agent answers calls in that voice. ElevenLabs bills you for the speech; we charge no credits for it.Carrier registration, opt-outs, and the gap between sent and delivered.
How to Send a Mass Text Message (SMS Blast) to Your Customers One text to a whole customer list, from Octo's chat or from the campaign editor: the message and its segment count, who gets it, what it costs before it goes out, and what the campaign page tells you afterwards. How to Get a Business Texting Number: 10DLC Registration or Your Own Twilio Number Why US carriers make a business register before it texts, the two ways to get a number that is cleared to send in Omni Octo (ask for one of ours, or bring your own Twilio number), what each texting state on Phone numbers means, and what you can get ready while you wait. How to Increase SMS Click-Through Rates: Tracked Short Links, Follow-Up Texts and Photo Replies Text messages have no open rate, so clicks and replies are what you can measure. Turn on tracked short links, see who clicked, send a second offer to the people who didn't, and let your AI agent answer the replies, photos included.
Connecting a bot, putting an AI agent on it, and the photos, voice notes and files customers send it.
What Telegram requires before your AI agent can answer On Telegram your business answers through a bot. What you need to make one, what its token is, how customers find it, and what Telegram does not let a bot do: write to someone first, or send a campaign. Connect your Telegram bot Paste your bot's token under Channels → Add Channel → Telegram. Omni Octo checks it, tells Telegram to send the bot's messages here with a secret only Telegram knows, and shows the bot on the channel page. What each message means, and how to change the token later. Put an AI agent on your Telegram bot Choose which agent answers your Telegram bot from the agent's Channels tab, or leave the bot to your team. What happens to messages with no agent, and what pauses the agent in a chat. Photos, voice notes and files on Telegram What your AI agent does with each thing a customer can send your Telegram bot: one answer to an album and its question, voice and video messages it listens to, GIFs and stickers, documents it cannot open, and where every file ends up in the inbox. Reply to Telegram customers from the inbox Take over a Telegram chat in /conversations, answer with text, a photo or a file, and hand it back to the AI. What the customer receives, what pauses the AI, and what you cannot do on Telegram. Your Telegram bot is not answering Messages do not reach the inbox, or they arrive and nobody replies. Where to look first, what each cause looks like on the channel page and in the conversation, and how to fix it.
Building one with Octo, what it is told, what it knows, and testing it before your customers do.
Build your first agent in one conversation Tell Octo what the agent is for, answer two or three questions, read its instructions, and say yes. Where to start, the one question Octo will not guess, what comes back, and where each piece lands on the agent page. Teach it what to say and how to behave The instructions an agent follows on every reply, the greeting it opens with, and the use case that can turn it into a questionnaire. Where each one lives on the agent page, how to write them so the agent follows them, and how to change them by asking Octo. Give it your business knowledge Crawl your website, upload the documents you already have, and paste what only lives in your head. Where the agent's knowledge base is, how to write it so the right passage is found, and how to keep it accurate. Test it before your customers do The test chat, the mic test, a call to your own phone, a call into the agent's number, and a simulated call you steer from Octo. What each one proves, what it needs, and the loop for fixing what you find. Put it to work Put the agent on your website, phone number, WhatsApp or email from its Channels tab, or by asking Octo. Who answers a channel and who can send from it, the disclosure callers hear, the on/off switch, and what to check when it isn't answering.
Outside systems that start the work — what they send, and what reaches your customer.
Shopify abandoned cart recovery, by phone call Someone leaves your checkout without paying. Later, your AI agent phones them by name — unless they bought in the meantime. Full setup, including the Shopify app, the permissions, and every place this goes quiet. Callback verification: text and call an investor about a held payment, and get a signed answer back Your payments system holds a suspicious payment. Omni Octo texts the investor, calls them, asks if they made the request, and sends a typed, signed result back to your system. A person still decides what happens to anything that is not a clear yes. Built from one message to Octo.
What every channel hands off to one inbox: alerts while the customer is still there, open requests, extracted fields, and handing a paused chat back to the AI.