Every inbound lead scored, explained and routed in seconds

A webhook receives the lead from your CRM or form. Qwen3 embeddings (via Ollama) compare it with your ideal customer profiles and past won/lost deals. Claude (or OpenAI) then qualifies it on BANT using that evidence, and the score, tier, next action and a reply draft are pushed back to HubSpot, Pipedrive, Slack or any webhook.

01Webhook inHubSpot · Pipedrive · forms

02NormalizeOne lead schema, PII-capped

03EmbedQwen3-Embedding via Ollama

04Vector fitICP similarity + kNN won/lost

05RulesEmail, title, spam, size

06LLM qualifyClaude / OpenAI, BANT JSON

07Blend & routeTier A–D + next action

08CRM outSigned webhook / API

Loading demo…

Run it for real

ollama pull qwen3-embedding:0.6b
cp .env.example .env (add an Anthropic or OpenAI key)
npm install && npm start

Connect your CRM

Point a HubSpot workflow, Pipedrive webhook or Zapier/Make/n8n step at POST /webhooks/lead/:source. Results come back as signed webhooks or direct CRM property updates.

Why vectors + LLM?

Embeddings ground the score in your history (what actually closed). The LLM reads intent and BANT and writes the reply. Rules add hard guards, so spam never reaches sales.