We debated the pricing model for weeks. Ultimately, we decided that visibility (Live Prospecting) shouldn't be gated behind a paywall. Small businesses need that data the most.

So, the core stack (Widget, Heat Scoring, Live Handoff, History) is $0. We only monetize via optional 'AI Power-Ups' (like Agent Coaching and Automated Outreach) for teams that are scaling.

We hope this aligns incentives: we only make money if you grow big enough to need automation.

A huge constraint we set was that Furie could not touch the host WordPress database. Too many chat plugins destroy PageSpeed scores by writing logs to wp_options or loading heavy assets.

We architected the plugin as a lightweight async wrapper. All the socket connections (Pusher) and state management happens on our external API layer. Even the chat history is stored remotely.

If you are a WP dev who avoids plugins for performance reasons, I'd love for you to audit our implementation and let me know if we hit the mark on performance.

We built Furie based on a frustration with the status quo: standard chat widgets are blind. You install them and wait for a 'bing.' Meanwhile, your highest-intent traffic (people reading your pricing page twice) leaves silently.

We tried to solve this by building a client-side 'Heat Score' algorithm. It monitors scroll depth, dwell time, and specific URL regex patterns to score purchase intent from 0-100 in real-time.

Ideally, this turns the chat widget into a radar for sales teams. I'd love feedback on the dashboard UI does the Heat Score give you enough signal to jump in manually?

While building the AI context engine, we realized that standard RAG (Vector Databases) was too slow for dynamic content. If a user is looking at a flash sale or a live inventory count, a cached vector embedding is useless.

We decided to architect Furie using Perplexity's Sonar models to live-crawl the specific URL the visitor is on during the chat. It added about ~400ms of latency compared to OpenAI alone, but the hallucination rate on pricing questions dropped to near zero.

Curious if other AI builders here have experimented with live-browsing models vs pre-indexed vectors?