Shipped build · Credit repair
Answers like a trained rep. Reads screenshots. Hears voice notes. Knows when to get a human.
Watch it work
What it does
- Answers customer questions to the firm's exact use case
- Keyword promos: "link" returns the booking link and the $25 consultation price
- Reads images — including credit-report screenshots — and replies in context
- Transcribes voice notes and responds to them
- Keeps full conversation history; hands off to a human when needed
- The client edits keywords, promo text, and pricing in a simple form. No developer required.
In the demo
A user uploads a credit-score screenshot. The agent reads it: "Your score is 500 — need it fixed for anything specific?"
A voice note arrives. The agent transcribes it and answers in context.
Much later, asked "what's my credit score?" — it answers 500. From memory, not a re-upload.
How it’s engineered
Human-paced chunks
Replies go out in human-paced chunks, never one block.
Selective persistence
The database stores useful information, not every message.
Scoped retrieval
Retrieval pulls only the messages a reply needs — smaller prompts, lower token cost.
Client-editable config
Consultation price, booking link, and promo codes are client-editable.
The system
Instagram DMs flow into the agent — vision, speech-to-text, thread memory — backed by a selective store with scoped retrieval, replying in human-paced chunks, with a human-handoff branch and a client-editable config form feeding keywords, pricing, and links.
Stack
- Instagram DM
- custom agent backend
- vision + speech-to-text
- retrieval memory
- client config form