Instagram Comments AnsweredIn The Brand's Own Voice.
A multi-tenant SaaS that connects to Instagram Business accounts, monitors incoming comments in real time, and generates brand-voice-aware replies. ZipLyne built the entire platform on Cloudflare, with an 8-layer spam gate that classifies every comment before spending a single AI credit.
The Actual Interface.
Every Comment Costs A Credit If You Let It.
An Instagram account of any size takes far more comments than a person can answer, and most of them are not worth answering. Sending each one to a language model means paying to classify spam.
ReplyMagic puts a rule-based gate in front of the model instead, and gives each merchant an isolated brand profile so the replies that do go out sound like them. Replies either auto-send or queue for human review.
Every Layer, From Scratch.
The 8-layer comment classifier. Bot detection, keyword filter, language detection, sentiment pre-screen, rate limit, block-list, pattern match, confidence threshold. Every comment is classified before it touches the LLM.
Tone extraction onboarding. Merchants paste examples of their ideal replies. The system extracts tone, style rules and vocabulary constraints, embedded into every generation prompt, so replies sound like the brand rather than a bot.
Dual reply mode. High-confidence comments fire immediately and edge cases route to a human approval queue. Merchants set their own confidence threshold per campaign: full control without manual overhead.
100% Cloudflare stack. Each merchant has isolated D1 tables, separate KV namespaces and a separate brand profile. Workers for compute, D1 for SQLite, KV for session and cache, R2 for media, and zero external infrastructure.
Rules Run First, So The Model Runs Rarely.
Generation. A fast, low-cost language model handles reply generation, with the brand voice system prompt injected on every call. It falls back to a secondary model automatically on rate-limit.
Classification. The rule-based layers run before any LLM call. Only comments that pass all 8 gates consume AI credits, which keeps cost per reply low at scale.
Reply analytics. Reply volume, approval rate, spam catch rate and cost per reply, surfaced per merchant in real time. Merchants know exactly what the system is doing and what it costs.
- Cloudflare Workers
- Cloudflare D1
- Cloudflare KV
- Cloudflare R2
- Language model
- Multi-model routing
- Instagram Graph API
Let’s Build What’s Next.
Bring the business problem. We’ll talk through what would make a difference and where to start.