Social Engagement
Drafts replies to comments and DMs in your voice, answers product questions from the brain, routes purchase intent, flags complaints to a human within the hour, and listens for mentions. Instagram Login API only. There is no compliant cold DM, so it does not do one.
Who it is for
The owner with 200 unanswered comments, and the marketer who knows the difference between a private reply to a commenter and a cold DM.
What it reads from the brain
Before every run. Sections are addressed as layer.section.
What it does, step by step
- Read the brain
Opens the sections above and the inputs.
- Gather evidence
Reads comments and the DM conversations customers started. Every finding cites its source.
- Decide and draft
Drafts a reply per comment and per open conversation in your voice, answers product questions from the brain, tags purchase intent and complaints. Uses only claims_allowed.
- Advise stops here
You get the queue with drafted replies and routing, and the escalations.
- Act
Replies to comments and continues DM conversations the customer started, within the platform windows, with the day's batch approved for the first 30 days.
- Write back
New objections and what did not work are written to the brain.
Tools it uses
instagram.comments.listComments on your postsinstagram.comments.replyApproval gatedinstagram.private_replyApproval gated. One per commenter within 7 days of their commentinstagram.messages.readConversations the customer startedinstagram.messages.sendApproval gated. Continues a conversation, never starts one
The three modes for this agent
Read tools only. You get the queue of comments and DMs with drafted replies and routing, and the complaints routed to you.
Replies and continued conversations go out after you approve the day's batch for the first 30 days. Act mode exists only after Meta App Review, which takes two to four weeks; we start that the day you say go.
Every hour. Same gates. Replies go out within the hour and complaints reach you first.
Guardrails and approvals
Enforced in code. Items with a filled circle wait for your tap in Telegram.
- Never argues.
- Never discusses refunds publicly.
- A complaint reaches a human within the hour.
- Daily action cap.
- One private reply per commenter within 7 days.
- No unsolicited DMs, ever.
- Meta App Review is a 2 to 4 week gate before Act mode exists.
- Every claim in every reply is in claims_allowed or is removed and listed as needs your confirmation.
What it writes back to the brain
Dated and sourced. Learnings and performance sections are append only.
Example output
Example output. Structure only. Counts are placeholders.
Social Engagement, your store, last hour
Ran at data level 1 (drafts only). Act mode waits on Meta App Review, 2 to 4 weeks.
- New comments: [n]. Replies drafted: [n]. Purchase intent tagged: [n], private reply drafted for each, one per commenter.
- Open DM conversations the customer started: [n]. Next message drafted for each.
- Complaints: [n], routed to you in escalations.md. No public reply drafted.
- New objection heard [n] times, written to the brain in the customer's words.
queue.csv (columns)
source | post_or_thread | customer_text | drafted_reply | routing | window_expires
escalations.md (one entry)
Comment on [post], [time]. Customer wrote: [their words]. Routing: complaint, to you, within the hour. No public reply from the agent.
Planned: a suggested private response for you to send yourself, attached to each escalation.
Also delivered: nothing else. The queue and the escalations are the whole output.