how it works

Five steps to replace your stack

No implementation team. No 30-day onboarding. Connect, train, configure, go live, pay for outcomes.

01

Connect your channels

OAuth into Shopify, Instagram, and TikTok in three minutes total. We pull catalog, products, recent orders, customer history, brand voice samples. No CSVs, no manual mapping.

02

Train your agents

Selify reads your existing knowledge — FAQs, returns policy, product docs, past support transcripts — and builds a per-tenant knowledge base. You can paste tone-of-voice examples or let it infer from your last 30 days of replies.

03

Set guardrails

Choose what agents can and can't do. Issue refunds up to $50 without escalation? Decline returns past 60 days automatically? Every action is policy-driven; nothing is silent.

04

Go live

Flip the switch per channel. Agents start handling messages, generating listings, syncing inventory. You see every action in a unified inbox; you can take over any conversation at any point.

05

Pay for outcomes

Each completed outcome (reply, try-on, listing, sync event) hits a meter. Tier subscription bundles a monthly quota; overage is pay-as-you-go. No seat fees, no per-bot fees, no surprises on the invoice.

stack

What powers it

layer tech why
Edge Cloudflare Workers + Pages Fast static + edge functions, low p95 latency.
Apps SvelteKit (dash + admin + selify.ai) SSR + island reactivity, tiny bundles.
Gateway Kong Single ingress, rate-limit, auth proxy.
Services FastAPI (Python) + axum (Rust) Python for AI, Rust for hot paths (auth, git).
Workflows Temporal (5 worker queues) Long-running jobs that survive deploys.
GPU Modal (DeepSeek + try-on + vision) Per-tenant inference, scale-to-zero.
Data Postgres + pgvector + RLS Per-tenant isolation, semantic search.
Auth Supabase Auth Standards-based JWT, OAuth providers.

enterprise customers can self-host the worker layer; control plane stays multi-tenant.