We make AI chatbots feel obvious.
knot is a small team building the AI infrastructure we wish existed: fast, multi-tenant, transparent, and copy-pasteable. No setup theatre. No vendor lock. Just answers.
Make grounded AI boringly easy.
The AI landscape is split into two extremes. On one side: raw API access — powerful, but you build the retrieval, the streaming, the tenancy, the analytics yourself. On the other: closed platforms that lock you into one model and one UI.
knot is the missing middle. The full RAG + multi-tenant stack, with the model layer still open. Bring your own key. Ship in an afternoon. Scale to a million conversations without rearchitecting.
"The visitor asks a question. The right answer streams back in under 300ms. Every other concern hides behind a sensible default."
Four opinions we'll never compromise on.
Ship the obvious thing.
Most AI products bury the magic under setup. We chase the shortest possible path from intent → answer.
Speed is a feature.
Sub-second streams aren't a luxury — they're the difference between a tool and a toy.
Tenancy is sacred.
Two layers of isolation, audited by default. Your data never bleeds into someone else's namespace.
Operator-grade UX.
We build for the engineer on Friday afternoon. Sane defaults, copy-pastable code, no surprises.
From weekend prototype to production AI.
Project knot started
Born out of frustration with bloated chatbot platforms. First prototype shipped in a weekend.
RAG pipeline goes live
Embeddings + vector retrieval + streaming LLM all stitched together. The grounded-answer loop closes in <300ms end-to-end.
Multi-tenant launch
Per-workspace isolation everywhere — Postgres, vectors, analytics. First paying customers go live.
BYO-key + Embed v2
Pro plan adds Bring-your-own provider keys. Widget rewritten in vanilla JS, ~12kb gzipped.
You're here
Public beta. Free tier forever. Self-serve onboarding. We're listening — talk to us anytime.
Built on battle-tested primitives.
We pick boring infrastructure so we can build interesting product.
sub-second time-to-first-token
swap providers without rewriting prompts
your data, isolated from every other tenant
row-level isolation on every write
incremental, re-runnable on every edit
no infra to babysit, scales to traffic
Small, opinionated, and shipping.
No middle management. Every line of code, every email, every onboarding call comes from someone on this list.
Ex-platform. Ships at the speed of the model.
Believes great UI is invisible.
Lives in embedding space.
Onboards every Pro customer personally.
Come build with us.
We're a small team. Every customer matters. Every bug gets a human.