An open-source AI back-office system for small businesses — built on markdown procedure files and the Claude API (Anthropic). No proprietary software, no subscription lock-in. You own the files. You own the data.
A 5-person plumbing company shouldn't need a $45,000/year back-office employee to chase invoices, manage scheduling, handle paperwork, and operate QuickBooks. But without that person, all of that falls on the owner — who is supposed to be running the trade.
This project is that back-office employee, built in AI.
The system is a library of markdown procedure files — plain text documents that tell an AI agent exactly how to do specific business jobs:
- Track projects and budgets
- Build estimates and proposals
- Generate and track invoices
- Monitor quality records
- Support business development
You load these files into a Claude agent (via the Anthropic API) and point it at your business data. The AI reads the procedures, follows the rules, and handles the back-office work. When your process changes, you edit a text file. No retraining. No DevOps.
QuickBooks is still your accounting system — the AI operates it on your behalf.
| Module | File | What It Does |
|---|---|---|
| Project Management | procedures/pm/pm-core.md |
Track jobs, hours, budgets, milestones. Proactive alerts when projects go off track. |
| Estimating — Core | procedures/estimating/estimating-core.md |
Build proposals from reusable workflow templates. No freeform hour guessing. |
| Workflow Library | procedures/estimating/workflow-library.md |
Library of reusable phase patterns with hour bands by role. |
| Invoicing | procedures/invoicing/invoicing-core.md |
Six billing models: percent-complete, milestone, deposit, T&M, fixed-price, materials-upfront. |
| Quality (ISO 9001-aligned) | procedures/quality/quality-core.md |
NCRs, CAPAs, document control, audit readiness. No certification required to use. |
| Business Development | procedures/bd/bd-core.md |
Capability statements, proposal content library, pipeline tracking. |
Schemas (JSON data structures the AI reads and writes):
opportunity.json, invoice.json, ratesets.json, quality-record.json, bd-content.json
- Markdown for logic, JSON for data. No proprietary formats. Any text editor can read and edit these files.
- The AI follows the procedure. All business rules live in the markdown files — not in code. Change the text, change the behavior.
- The owner approves before anything goes to a client. Invoices, proposals, follow-up emails — the AI drafts, you approve.
- Your data is always readable without the platform running. Key records export to your Google Drive nightly as PDFs and CSVs. If anything breaks, your data is right there.
- QuickBooks-first. Don't rebuild what QuickBooks already does. Integrate with it.
Tier 1 — Service businesses (plumbers, electricians, HVAC, catering, cleaning, landscaping)
CRM, scheduling, project tracking, invoicing, inventory, AI chat assistant
Tier 2 — Small manufacturers and ISO-pursuing firms
Everything in Tier 1 + ISO 9001, CAPA, document control, supplier management, BOM
The framework is generic — configure it for your business by filling in your rates, your team, your workflows.
This is an early-stage, work-in-progress framework. The procedure library is functional and covers the core back-office modules for a service business or small engineering firm. Wiring it up to live APIs (QuickBooks, Google Workspace) is a next phase.
What exists today:
- Complete procedure file library for core modules
- JSON schemas for all major data structures
- Architecture validated against commercial alternatives
What doesn't exist yet:
- Runnable agent code
- QuickBooks API integration
- Google Workspace integration
- A UI
If you are a developer, consultant, or technical business owner who wants to try this or build on it — that's exactly who this is for right now.
You need an Anthropic API key. Claude Sonnet is the recommended model.
Basic approach:
- Clone this repo
- Pick the procedure files relevant to your business
- Load them into a Claude agent as system prompt context
- Point the agent at your project/client data (JSON files in Google Drive or similar)
- Start asking it to do back-office tasks
A proper setup guide is on the roadmap. In the meantime, the procedure files are readable — they describe exactly what the AI should do and what data it needs.
This is a public experiment. If you try it, find a gap, think a procedure is wrong, or have a use case that isn't covered:
Useful feedback:
- "This doesn't cover [business type] because..."
- "The invoicing procedure breaks for [billing scenario]..."
- "I tried loading this into [tool] and ran into..."
- "A plumber would never [describe assumption]..."
This repo covers commercial and service businesses. If your business pursues government contracts — SAM.gov, SBIR, RFPs, ITAR, CMMC — that's a separate, more complex system. Reach out directly if you want to talk through it:
Mike Maier — github.com/mminwi
MIT. Use it, modify it, build on it. See LICENSE.
Built by Mike Maier.
The framework is open source — use it, modify it, build on it under the MIT license. If you want help implementing it for your business, open an issue or reach out directly.