What works: the AI answers Instagram Direct messages with the same prompt, knowledge base and tools as WhatsApp/Telegram (or a dedicated Digital Employee assigned to the Instagram channel), and conversations appear in Chats, History and contact/lead timelines. On top of that, the Comment→DM funnel turns public comments into private AI conversations (below).

Connect it — the easy way (1-click): on Integrations → Instagram click Connect Instagram, sign in with your Instagram Business/Creator account and grant access — we store the token, map the account and subscribe to webhooks automatically.

Or set it up manually (advanced): add your Instagram Business Account ID (mapping), paste your access token + app secret in the credentials panel (masked once saved), and set the webhook URL + verify token in the Meta console subscribed to messages + comments.

Good to know:

• DMs are text-based; the public comment reply is one short line by design (specifics go to DM).
• You can reply to Instagram DMs from the dashboard Chats page and from the SIPmind app within Meta's 24-hour window (outside it, reply from the Instagram app); incoming messages also forward to your operator group, and a reply typed in the group goes back to the person in Instagram.
• Reply from the Instagram app and the assistant steps back. The moment you answer someone yourself in Instagram, the assistant stops answering that person for five minutes (every message they send meanwhile extends the pause), your message appears in their thread in Chats, and the assistant sees it when it takes over again. Nothing to switch — it is the same rule as answering from Chats or the operator group.
• Several messages in a row get one answer. People write in bubbles; the assistant waits a few seconds for the burst to finish and answers everything at once instead of replying to each bubble separately.
• "Can I talk to a real person?" The assistant does not ask for a phone number or email — on Instagram the chat is the contact. It tells you (operator group, otherwise email), says so to the person in one sentence, and the thread stays in "Needs reply" until you answer. Until then it keeps its replies short and does not start the pitch over, and it says "the team will answer here" once — a bare "ok", "waiting" or "?" after that gets no reply from the assistant (the message still reaches your operator group), while anything new the person writes is acknowledged in a few words. If they are still waiting after ten minutes, the assistant reminds your group that someone is waiting, at most every twenty minutes, and tells the person so. This works on WhatsApp and Telegram the same way.
• Comment auto-reply and the DM funnel are opt-in per account and OFF by default.
• Keyword triggers replace the automatic lead detection. Leave the field empty and the AI DMs anyone whose comment reads like interest. Fill it in and only comments containing one of those words start the funnel — a "how much is it?" without the keyword will get a public reply but no DM. Matching ignores case and works inside a longer comment, but not across spaces: spring2026 will not match "spring 2026". A keyword also outranks the spam filter, so a bare code word is never mistaken for spam and quietly dropped.
• If Instagram stays silent right after connecting, look for an amber note in Integrations → Instagram. Linking the account and subscribing it to events are two separate steps on Meta's side, and the second one can be refused — most often because the account does not have access to our app yet. While that note is on the page, no message or comment reaches us; press Retry on it. The note disappears by itself once Meta confirms.
• Instagram access renews itself. Meta expires the connection 60 days after you connect it; we renew it automatically well before that, and Integrations → Instagram shows how long it is valid for. If a renewal ever fails, that line turns red and your operator gets a Telegram message — reconnect with the same "Connect Instagram" button.
• See what the AI did under your posts. Every comment it acted on is recorded on the History page — use the IG comments chip to show only those. Each row carries the comment text, how the AI read it, and whether a DM was opened; a comment the AI hid is marked hidden. Spam that was merely ignored is not recorded — nothing was done to it.
• One Save, whatever you touched. Each scenario row has its own Save, and the button at the bottom of the section saves the switches. If you edited rows and pressed the bottom button, it now saves those rows first and the switches after — a row you edited is marked "not saved" until it is. If one row cannot be saved, nothing is saved and the page says which, with your text still on screen.
• The two fields answer two different questions. On the left, WHO: empty and the agent decides for itself who sounds interested; filled and only those words fire it. On the right, WHAT: empty and the agent answers in its own words; filled and exactly that text goes out, letter for letter. A link or a promo code always belongs on the right — a model may rewrite it. The default scenario usually wants both empty: nothing specific was promised there.
• An empty reply hands the tap to the agent. The reply field is for something that must arrive exactly as written — a link, a promo code, an address. Leave it empty and the tap is answered by the agent from your prompt instead, as if the person had said the button's own words. That is what the default scenario usually wants: nothing specific was promised, so a real answer beats a canned one.
• Pause keeps everything, stops everything. Pause next to the account leaves the token and every scenario in place, but the AI stops reacting to comments and DMs on that account until you press Resume. Nothing is deleted.
• Disconnecting really disconnects. Disconnect next to the account deletes the stored access token, unsubscribes us from Meta's webhooks, and removes the account link — we stop receiving that account's messages and comments. Pause (above) is the softer option: it keeps the connection so you can switch it back on.