zambo.dev — 19+ products. ask your agent about it.
Tags: webdev, AI, showdev, opensource
I've been building alone for the past few years. No co-founder, no funding, no team. Just TypeScript, a lot of coffee, and a habit of shipping things before I talk about them.
The result is zambo.dev — 19 live products across SaaS, AI agents, onchain tools, and dev infrastructure.
This post isn't a launch announcement. It's a breakdown of how I built it, why it's structured the way it is, and the one thing I did recently that I think more solo builders should be doing.
The stack
Everything runs in a pnpm monorepo on Node 24 + TypeScript 5.9:
API: Express 5, base path /api, single server for all products
DB: PostgreSQL + Drizzle ORM
AI: Groq 4-model cascade — llama-3.1-70b → 8b-instant → gemma2 → mixtral → hardcoded fallback. Users never see an error.
Payments: PayPal Subscriptions (auto-creates products and plans from credentials on first run)
Email: Resend for transactional + drip sequences
Frontend: React + Vite + shadcn/ui + Framer Motion
The products (short version)
ZAMBRO — paste any URL, startup idea, wallet address, or GitHub repo. Get an opportunity audit in under 5 seconds. Powered by the same Groq cascade.
ZAMBOT — AI agent evolution engine. POST a goal, get back a strategy, task list, and agent config. 3 free sparks/day, then 0.001 USDC on Base via x402. No accounts. No API keys.
LeadSignal — AI lead generation for contractors. 24 trades × unlimited cities. Free tier, then $49–$199/mo via PayPal subscriptions.
X711 — universal MCP server hub. Connect any AI agent to 50+ tools (GitHub, Linear, Notion, Postgres, Stripe) through one endpoint. No per-tool auth setup.
Signal — verified professional passport. On-chain proof of work for developers and builders.
ProvibeCode — AI code audit for vibe-coded projects. Catches the stuff LLMs miss when they write their own code.
CreditHunt — 29 verified funding programs, $813K+ in available capital, refreshed daily.
Plus: Criptic, BountyLayer, Govri, VeriAI, Dunnly, Entangler, ProofForge, Substrate Layer, AgentWorld, Dunnly.
The architecture decision I'm most glad I made
One unified API endpoint:
POST https://zambo.dev/api/zambo
Content-Type: application/json
{ "need": "I need contractor leads in Phoenix for HVAC" }
Response:
{
"understood_need": "AI lead generation for HVAC contractors in Phoenix, AZ",
"solution": "...",
"powered_by": ["LeadSignal", "ZAMBRO"],
"next_steps": [...],
"call_to_action": "Run LeadSignal for HVAC + Phoenix now — first 3 searches free",
"tier": "free",
"meta": { "requests_remaining": 9, "reset_in_hours": 24 }
}
It routes through a classifier, pulls from the relevant products, runs Groq, and returns structured JSON. Free, no key, no signup. 10 requests/day per IP.
This single endpoint is what made the next decision obvious.
Making the whole stack AI-queryable
About a week ago I realized something: every AI agent a developer uses — Claude, Cursor, Perplexity, GPT — can already browse or fetch context from URLs. If I structure my site correctly, any agent can brief a new user on every product, return integration docs, and route them to the right tool automatically.
So I built the GEO layer:
zambo.dev/llms.txt — full LLM context file, every product, every endpoint, every integration pattern
zambo.dev/agent.json — machine-readable manifest with capabilities, pricing, auth requirements
zambo.dev/ai.json — structured entity catalog
zambo.dev/.well-known/agent.json — standard well-known path for agent discovery
POST zambo.dev/api/geo/ping — telemetry endpoint for agent traffic
Now when someone asks Claude "how do I integrate ZAMBOT into my agent?" — it returns the actual integration code from my llms.txt context, correct and up to date.
I added a hero to the homepage that shows this live: Claude, Cursor, and Perplexity each typing real queries and streaming back real answers — pulled directly from the indexed content.
The typewriter effect is real. The responses are accurate. The endpoints it references actually work.
What solo building actually looks like
The honest version:
Every product shares one Express server. Adding a new product is: new route file, mount it in index.ts, done.
The 4-model AI cascade means I never have to worry about a single model going down or rate-limiting. It just falls through.
PayPal plans auto-create on first subscription attempt. I never touch the dashboard.
Email deduplication is built in at the DB level — hasEmailBeenSent(email, type, hoursWindow) runs before every Resend call.
The drip engine runs every 30 minutes via a setInterval on boot. No cron, no external scheduler.
None of this is elegant. It's just practical. When you're solo, "it works and I understand it" beats "it's architecturally pure" every time.
Try it
curl -X POST https://zambo.dev/api/zambo \
-H "Content-Type: application/json" \
-d '{"need": "audit my GitHub repo for production readiness"}'
Or ask Claude: "tell me about zambo.dev"
It knows.
→ zambo.dev
Built by Brennan Zambo — @Cripdoecrypto
Top comments (1)
happy to answer any questions on telegram as well. @cripticweb3