knot.ai
Platform

Built like the AI products you actually want to use.

Six core capabilities engineered for production from day one — strict retrieval grounding, real operational analytics, surgical multi-tenant isolation, and provider flexibility.

RAG Knowledge System

Retrieval-grounded answers, not improvisation.

Every chat is anchored to your data. We chunk your knowledge with sentence-boundary awareness, semantically embed each chunk, and store vectors in a dedicated namespace per chatbot — strict tenant isolation at the vector layer.

  • Smart chunking with sentence-boundary overlap
  • Semantic embeddings tuned for retrieval recall
  • Top-K semantic search at query time (configurable)
  • Strict 'no fabrication' system prompt — falls back if unsure
RAG flow
Question
Are you open on Sundays?
Embed → top-K
Vector search returns 3 chunks · cosine 0.84 · 0.79 · 0.71
Grounded answer
We're open Monday-Saturday, closed Sundays. Emergency support is available 24/7 through partner hospitals.
Analytics Engine

Operational metrics, not vanity numbers.

We track what actually tells you whether your bot is working: per-bot fallback rate, latency p50/p95, hour-of-day activity, embedded-site breakdowns, conversation transcripts with replay. Built for product teams who measure outcomes.

  • Fallback rate per chatbot — the #1 'is it useful?' signal
  • Latency histograms with p50 / p95
  • Per-embedded-site usage breakdown
  • Full conversation logs with source URL + transcripts
Fallback
4.2%
p95
342 ms
AI Routing

Pick the model. Bring your key. Switch live.

Provider-agnostic dispatcher. Multiple leading LLM providers supported out of the box, each with a curated model list. Bring your own provider key — we validate it live before saving and the chats route through your account instead of ours.

  • Multi-provider model routing
  • Live API-key validation (test before saving)
  • Tone, temperature, max-tokens controls
  • Per-chatbot fallback message — 100% enforced
Model
Fast streaming
Default · sub-second TTFT
Active
Reasoning
Step-by-step analysis
Long context
Wide window for documents
Embed System

One snippet. Any framework.

Vanilla-JS widget, zero dependencies, themed from your dashboard. Drop a <script> tag into HTML, paste a React component into JSX, or use the iframe variant for pinned in-page placement. Every load registers an origin so you see exactly which sites are using the bot.

  • Script tag, React component, or iframe — your choice
  • Domain whitelist + per-origin activate/deactivate toggle
  • Auto-themes with your brand colors
  • Heartbeat tracking — know when a site removes the snippet
Embed snippet
<script
  src="https://knot.ai/widget.js"
  data-chatbot-id="bot_..."
  data-api="https://api.knot.ai"
  defer
></script>
Multi-Tenant Architecture

Built for SaaS scale on day one.

Two layers of tenant isolation by design. Row-level isolation on every database table scopes data by user identity. Each chatbot gets its own vector namespace — vectors can't bleed across tenants even at the storage layer.

  • Row-level isolation on chatbots, knowledge, conversations, messages, embeds
  • Per-chatbot vector namespace — strict semantic isolation
  • Server-side ownership re-verification on every write endpoint
  • Admin credentials never leave the backend — public-side key is safe for the browser
Isolation
Database
Row-level isolation · scoped by user identity
Vectors
Namespace per chatbot · vectors can't leak
API
Admin credentials server-only · ownership re-checked
Enterprise Security

Audited boundaries. Yours alone.

Managed Auth with OAuth + magic-link, encrypted at rest, JWT verified on every backend call. Plan-aware limit enforcement at every write endpoint — chatbots, knowledge, embedded sites, monthly messages — all gated server-side, never trusted from the client.

  • JWT verified on every protected route
  • Plan limits enforced server-side (chatbots, knowledge, sites, messages)
  • Per-origin deactivation — kill an embedded chatbot instantly
  • Bring-your-own LLM key — your tokens never proxy through us
Limit guards
Messages / mo184 / 10,000
Chatbots3 / 5
Embed sites2 / 5
Enforced server-side at every write endpoint

Try every feature, free.

One chatbot, 1,000 messages a month, no card. Upgrade only when you need volume.