I shipped 6 open-source AI tools for small businesses in 30 days

Steff D — Wed, 20 May 2026 23:44:41 +0000

Why I'm posting this

Most small-business owners try ChatGPT, get generic answers, churn within 60 days. Custom AI consulting starts at $5K-$15K. There's a real gap in the middle.

For the last 30 days I've been building toward closing it. Every artifact below is open source, MIT licensed, free to use right now — no signup, no email gate.

The 6 tools

1. CLAWVR MCP Server — install once in Claude Desktop, get 4 tools for generating SMB AI strategies across 13 verticals.
→ https://github.com/Steffd415/clawvr-mcp-server

2. CLAWVR Agent Skill — drop-in Claude Skill that auto-detects your business vertical and outputs a custom AI Operating System.
→ https://github.com/Steffd415/clawvr-agent-skill

3. 25 vertical AI playbooks — master prompts + 6 workflow superprompts each for dental, contractor, restaurant, real estate, law, vet, salon, photographer, fitness, insurance, CPA, pet groomer, auto repair, tutoring, lawn care, and more.
→ https://gist.github.com/Steffd415

4. Master prompts corpus — the canonical repo aggregating everything above.
→ https://github.com/Steffd415/awesome-claude-prompts-for-smb

5. Free AI Business Audit Tool — 8 questions, ~90 seconds, Gemini-generated 1-page strategy custom to a single business.
→ https://clawvr.com/free-audit

6. 6 Custom GPTs on the OpenAI GPT Store — vertical-specific GPTs (real estate, CPA, lawyer, dental, HVAC + a master strategy generator).
→ https://chatgpt.com/g/g-6a0c95e1f47081918770136430cfd3bb

All 6 in one place: https://clawvr.com/tools

Why open source

The paid version of this (custom delivery with 18 superprompts + a 30-day roadmap + ROI projections) is $297 one-time. The free tools above will get you 80% of the value if you do the setup yourself.

For everyone who can't, the paid version exists. For everyone else, the open-source corpus is yours.

Tech stack (for the devs)

Next.js 16 App Router on Vercel Fluid Compute
Supabase Postgres + RLS
Gemini Flash 2.5 for generation (~$0.001 per audit)
MCP server: TypeScript + @modelcontextprotocol/sdk + Zod
Claude Agent Skill: just SKILL.md + vertical templates (no code)
Self-hosted email tracking (open-pixel gif + 302 click redirect) — built it tonight because Instantly's webhook is paid-tier

Who this is for

Devs building tools for small businesses: these patterns might save you time
SMB owners: install the MCP server in Claude Desktop tonight and watch what happens
Makers: the entire open-source corpus is yours to fork, reuse, or improve

Looking for feedback, contributions, or rebuttals. What would you ship next?

— The CLAWVR team
hello@getclawvr.com | https://clawvr.com/tools

Why most small businesses quit ChatGPT in 60 days (and the 1-paragraph fix that changes everything)

Steff D — Tue, 19 May 2026 17:19:37 +0000

McKinsey shipped a study last year that nobody talks about:

72% of small businesses tried generative AI in 2024
Only 19% were still active weekly users 6 months later

Nearly three out of four SMB owners quit. Why?

Every quitter says the same thing: "AI doesn't work for my industry."

That's almost never true. The AI works fine. The problem is the input.

The pattern I see daily working with SMB owners

They sign up for ChatGPT, type "write a follow-up email," get something generic and robotic, and conclude AI doesn't work for their business.

The model never had a chance. It had zero context about:

Their sub-vertical (boutique dental vs corporate dental)
Their software stack (HubSpot vs Mailchimp vs nothing)
Their customer profile (cash-pay luxury vs insurance-driven volume)
Their voice (warm-and-folksy vs clinical-and-precise)
Their typical ticket size ($500 vs $50,000)
Their compliance constraints (HIPAA, GDPR, FINRA)

Without that context, every output is generic. Generic outputs feel useless. Owner quits.

The 1-paragraph fix

Before you ask AI to do anything, paste this paragraph into a fresh chat (or into the System Prompt / Custom Instructions field):

You are the senior operations partner for [BUSINESS NAME], a [SUB-VERTICAL] business located in [CITY]. We have [TEAM SIZE] employees. Our typical customer is [CUSTOMER PROFILE]. Our ticket size is [PRICE RANGE]. We use [SOFTWARE STACK] day-to-day. Our voice is [VOICE DESCRIPTION — e.g., warm but direct, technical but approachable]. The 3 biggest weekly time-sinks we want to automate are: [PAIN 1], [PAIN 2], [PAIN 3]. Treat every reply as if you have been with the company for 6 months and know our business intimately. Default to specifics over generalities. Push back on vague requests by asking 1 sharp question before drafting.

That's it. Fill in the brackets, paste it, then ask whatever you were going to ask.

The difference is immediate. Most owners' jaws drop the first time.

Why this works (the technical part)

LLMs are next-token predictors. Their outputs are conditioned on whatever context is in the conversation. With zero business context, the model defaults to the average of its training data — which is a soup of every business that ever existed. The output is the average of everything, which is generic.

The master prompt loads your specific context. Now the model's predictions are conditioned on YOU. The output sounds like your business because it's mathematically anchored to your business.

What to do after the master prompt is loaded

Layer workflow-specific superprompts on top. Each one inherits the master context. For example:

Follow-up email superprompt:

Write a follow-up email to a prospect who [SPECIFIC SITUATION]. They previously said [OBJECTION OR INTEREST]. Reference 2 specific things from our last conversation. Match our voice. Keep it under 120 words. End with a single specific ask.

Review response superprompt:

Write a response to a [STAR RATING]-star review that says [REVIEW TEXT]. Acknowledge the specific issue. Reference our brand promise around [VALUE]. Offer one concrete next step. Don't be defensive. 80-100 words.

Now you have 18 of these prompts (one for every workflow your business actually runs), all inheriting the master context. That's an AI Operating System.

The productized version

Writing all 18 superprompts yourself takes ~6 hours if you know what you're doing. Most owners don't.

That's what we built at clawvr.com. 12-question intake, we generate the master prompt + 18 industry-specific superprompts pre-loaded with your business context. PDF in 5-10 minutes. $297 one-time. No subscription. Buyer pastes the prompts into their own ChatGPT / Claude / Gemini account.

But even without paying for the productized version, the 1-paragraph fix above is the single highest-leverage change most SMB owners can make to their AI workflow this week.

Try it. The first time you see the output difference, you'll understand why 72% of owners quit — they never loaded the context.

The CLAWVR team. We build custom AI Operating Systems for small businesses. clawvr.com

DEV Community: Steff D