Canary Chat
AI Chatbot That Learns Your Business
I've been building Canary for the past two months. It's an AI chatbot that businesses embed on their website -- it learns from their content, answers visitor questions, captures leads, and hands off to humans when it's stuck.
The idea came from a simple observation: Intercom charges $39/seat/month minimum. Drift starts at $2,500/month. For a small business that just wants "answer questions on my website and collect emails," that's absurd.
What it does:
- Paste your website URL, the AI scrapes and learns your content in under 5 minutes
- Visitors get instant answers grounded in your actual website content (not hallucinated)
- Lead capture built in -- name, email, phone collected naturally during conversation
- Human handoff when the AI can't answer
- Analytics dashboard showing conversations, leads, conversion rates
- WhatsApp channel support
- Custom disclaimers, follow-up questions, proactive messages
The stack (for the technical folks):
- Express + TypeScript API on Render ($7/mo)
- React admin dashboard on Vercel (free)
- Preact widget (4KB) on Vercel (free)
- Next.js landing page on Vercel (free)
- Supabase for data ($25/mo)
- OpenAI GPT-5.4-nano for responses (~$50/mo for 10 tenants)
- Firecrawl for website scraping ($16/mo)
- Clerk for auth (free tier)
- Resend for transactional email (free tier)
Total infra: ~$127/month running 10 active tenants. Each tenant gets their own isolated knowledge base, custom branding, and analytics.
What I learned:
1. Smart onboarding matters more than features. I use Firecrawl's map endpoint to discover all URLs on a site, then an LLM picks the 5 most important pages to scrape. Onboarding went from 10 minutes to under 60 seconds. That single change doubled completion rates.
2. The system prompt is 80% of the product. I spent more time tuning the system prompt than building features. Response format rules, grounding instructions, when to say "I don't know" vs. when to attempt an answer -- this is where chatbot quality lives.
3. Per-resolution pricing is the wrong model for SMBs. Intercom charges $0.99 per AI resolution. That means your bill goes up as the AI gets better. I went with flat monthly pricing instead. Predictable for the customer, simpler for me.
Current state:
- 4 active tenants (2 paying clients, 1 internal, 1 production demo)
- Free tier: 50 conversations/month
- Growth: $49/month
- Scale: $149/month
- Built solo, bootstrapped
The site is live at canarychat.app if anyone wants to try it. You can paste any URL and chat with your site's content in the free tier.
Happy to answer questions about the architecture, pricing decisions, or anything else. Building in public so nothing is off limits.
1 Comment
Disrupting Intercom and Drift on price is a smart play for the SMB market, especially with that $127/mo infra efficiency. But here is the hurdle you’re going to hit: Perceived Value.
In the SaaS world, when something is significantly cheaper, businesses often equate it with 'higher risk.' For a business to put Canary on their front page and trust it with their leads, they need more than just a low price—they need to know you are a legitimate player.
If Canary is being featured in top-tier tech or business media as the 'Intercom Killer' for SMBs, that 'Sasta' label instantly turns into 'Smart Choice.' Without that high-level media validation, you're just another bot platform in a crowded market. Have you thought about how a feature in a major publication could bridge that trust gap and turn your low price point into a position of authority?