Tenant isolation at the query level
Every retrieval filters by tenant_id at the database layer, with row-level security as a backstop. Cross-tenant leakage is not a policy — it is mathematically impossible.
Our mission
Make AI support feel human.
Most support AI is a thin wrapper over a chat completion. Zybo AI is the layer underneath — hybrid retrieval, confidence scoring, a real agent inbox, and tenant-strict isolation, built as one system rather than five glued together.
We believe an AI that knows when to stop guessing is more valuable than one that always has an answer. So we built the platform around that idea: confident replies when the docs back them up, a fast handoff to a human when they do not.
AI when it can. Humans when it matters.
How we build
Principles we ship by.
Engineering principles taken straight from the codebase, not the marketing site. Each one shapes a real constraint somewhere in the platform.
Confident answers, honest deflection
Every AI response carries a confidence score. Below the configured threshold (0.3 by default) the AI agent escalates to a human instead of guessing. Wrong answers cost more than slow ones.
Real-time everything
Streaming replies, typing indicators, agent presence, and queue state flow through Socket.IO with rooms scoped per conversation and per organization. The UI never has to poll.
Production-ready from day one
pgvector for embeddings, BullMQ for ingestion and summarization, daily-rotating Winston logs, idempotent migrations, and an output filter on every LLM response. No staging-only shortcuts.
Auditable by design
A global mutation interceptor and auth-event instrumentation write to an append-only audit log per workspace. Every change has an actor, an IP, and a timestamp you can query.
No hardcoded business logic
Plan limits, feature flags, rate limits, prompts, and operation modes live in the database — not buried in service code. New features ship as data, not deploys.
Parent company
A Zyfolks Technologies product.
Zybo AI is built by Zyfolks Technologies — a software studio that ships custom AI, integration, and product engineering work for teams who need more than an off-the-shelf SaaS.
The same team that designs the systems behind Zybo AI runs Zyfolks day-to-day. That means production-grade engineering, in-house ML expertise, and a roadmap shaped by what we see going live in the field.
Visit zyfolks.comStack
Built with tools we trust.
A short list, picked for boring stability and well-understood failure modes. We try not to add a dependency unless we would also want to maintain it.
- TypeScript strict mode, end to end
- NestJS modular API, per-feature modules
- PostgreSQL + pgvector hybrid search, HNSW index
- Redis + BullMQ rate limits, workers, session memory
- Socket.IO streaming + real-time inbox
- Astro this site you are reading
- Tailwind design tokens, light + dark
- nginx static delivery + edge caching
Want to talk to the team ?
We answer fast, in plain English, and we read every message. Tell us what you are trying to ship.