From SAP Consultant to Open Source ERP Builder: Why I Built ERPClaw

In 2009, in a fluorescent-lit conference room in Greensburg, Pennsylvania, I watched a $200-an-hour SAP consultant explain the Vendor Master configuration screen to an Allegheny Power accountant who had been doing accounts payable for thirty-one years.

She already knew the answer. She had known it for a decade. She was waiting, politely, for the consultant to finish so she could nod and go back to her desk and approve the invoice that had been sitting in her queue for two days because the workflow we had configured was wrong.

I was the consultant’s junior. My billable rate was lower. I was twenty-four years old and I had been at Accenture for less than a year. I remember thinking, very clearly, two things at once. First, that this was the most expensive way to discover something that the person sitting across the table already knew. Second, that I was going to do this for at least the next five years because the project plan said so.

That moment is the reason ERPClaw exists. It just took seventeen years to get from there to here.

The Accenture years

I spent six years at Accenture rolling out SAP for utility companies. Allegheny Power. E.ON in the UK and Germany. American Water. The work was good. The clients were serious people running real infrastructure that millions of households depend on. I learned more about the actual mechanics of how a large business runs in those six years than I have learned in any other six-year window of my life.

I also learned what enterprise software costs.

The published license fee for an SAP S/4HANA seat is one number. The real cost is something between three and ten times that number, depending on who you ask. The implementation partner bills hours. The change management team bills hours. The training team bills hours. The integration team bills hours to talk to the data team that bills hours to talk to the security team. A mid-sized utility writes checks for tens of millions of dollars before a single accountant can post a journal entry.

I am not going to pretend SAP is bad software. It is not. It runs the spine of global commerce and it has earned that position. What I will say is this: the gap between what SAP actually does for a business and what the business pays to get it has been growing for thirty years, and nobody on the buying side has had a real alternative.

The thing the consultant was explaining to the accountant in Greensburg was not complicated. It was a screen with maybe twelve fields. The reason it was expensive was not the screen. It was the cost of getting the screen, configured correctly, in front of her, in a system that her company had committed to for the next fifteen years.

I started writing code on the side around year four. Mostly utilities. A small reporting tool. A spreadsheet macro that saved a client team six hours a week. Nothing serious. But the contrast was sharp. Six hours of consulting time was around $1,200. The macro was an afternoon and it ran forever for free.

I started thinking about what it would take to build software that did the work of an SAP module without the implementation overhead. I did not have an answer in 2012. I did not have one in 2014, either. I left Accenture in 2015 and started building.

The indie SaaS years

Between Accenture and ERPClaw there is a decade I do not talk about much. It is not a secret. It just gets compressed in my head into a single phrase: “I was building inventory software.”

I built ZapInventory because I kept getting asked by small wholesalers and Shopify merchants to help them figure out their stock. They were on QuickBooks. QuickBooks does not really do inventory. They were on spreadsheets. Spreadsheets do not really do inventory either. They needed something in between QuickBooks and a $50,000 NetSuite implementation, and almost nothing existed in that gap.

ZapInventory grew. By 2020, Crozdesk ranked it the third inventory management software in the world, ahead of SAP Business One and Zoho Inventory. That sentence still feels slightly absurd to type. I was a solo founder running a small team out of Pune. We were ahead of SAP on a list that mattered to buyers. Not because we were better than SAP. We were not. We were ahead because for the buyer who came to that list, we were the right thing and SAP was the wrong thing, and the list was honest about that.

ZapInventory was acquired by InvenSync Inc. I built a few other things. StockVR, a 3D warehouse visualization tool, got mentioned by Allied Market Research in 2024 in the same paragraph as Oracle, IBM, and SAP. SuperSnail, a mobile game my team made on the side, crossed a million downloads and got acquired by QCPlay Digital. I became an IEEE Senior Member somewhere along the way, published a couple of papers in Springer proceedings on AR and inventory visualization. The CV grew. The lessons compounded.

The lesson that stuck, the one that mattered for what came next: the default ERP for a small business should not cost $50,000 to install. It should cost approximately zero, and it should know what it is doing.

In 2024 I did not yet think that was possible. By the end of that year I started to think it was.

The thing that changed

Two things changed.

The first was AI. Not the chatbot kind. The “I can describe what I want and the software writes itself” kind. By late 2024 I was writing more code per week with AI than I had written by hand in the previous three years combined. The cost of building software, on the maker side, fell off a cliff. The cost of buying software, on the buyer side, did not.

That gap is an arbitrage. It is the kind of arbitrage that does not last. Either buyers stop paying or sellers start cutting prices. Both are happening, slowly, in pockets. ERPClaw is a bet that the cleanest way through the arbitrage is to build the new default from scratch, AI-native, free, open source, and let the gap collapse on its own.

The second thing that changed was OpenClaw. OpenClaw is an AI-native skill execution layer. You install a skill, you talk to your assistant in natural language, the assistant calls the right action under the hood. It is not a chatbot bolted onto a CRUD app. It is a runtime where the skill is the unit of distribution and the AI is how the user interacts with it. Once OpenClaw existed, the architecture for what I had been wanting to build for a decade clicked into place in a weekend.

ERPClaw is the answer to the question I have been carrying since that conference room in Greensburg. What if the accountant just talked to the system? What if the system did not require a $200-an-hour consultant to explain a screen? What if the screen, in fact, did not really need to exist?

What ERPClaw actually is

In concrete terms, ERPClaw is a complete double-entry ERP that installs in one command, runs locally on a SQLite database, and is driven by natural language through your AI assistant.

It has 48 modules. Core accounting, sales, purchasing, inventory, HR, payroll, manufacturing, projects, assets, CRM. Vertical packs for healthcare, education, property management, retail, construction, food service, hospitality, legal, nonprofit, and a dozen others. Deep integrations with Stripe and Shopify that post per-transaction journal entries directly into your general ledger. ASC 606 revenue recognition. ASC 842 lease accounting. W-2 generation. Multi-entity consolidation.

It is MIT licensed. It is free. It runs on your laptop or your own server. Your data lives on disk in a single file you own. The full source is on GitHub. If you want to fork it, fork it. If you want to install it, the docs are public.

There is also a self-extending layer called ERPClaw OS. It is the part of the system that lets AI agents generate new vertical modules within constitutional financial laws. We have used it to ship three modules end-to-end with no human-written code: groomingclaw, tattooclaw, storageclaw. They pass the same 7,627-test suite the hand-written modules pass. The constitution does not let them post bad GL entries. That is the part of the architecture I am proudest of, and it is the part I think will matter most over the next five years.

You can see the full feature set at /features, and the pricing is on /pricing. The pricing page is short. The product is free. The hosted version, when it ships later this year, will be priced at the cost of running it for you, with a generous free tier.

The bet

The bet I am making with ERPClaw is that AI-native is not a feature. It is the new default architecture for business software, and the next ten years are going to look like the migration from on-premise to SaaS that happened between 2008 and 2018. Same shape, faster timeline.

A SaaS ERP that bolts a chatbot onto a 1995 data model is not the answer. The answer is a system designed, from the database layer up, to be driven by an AI agent on behalf of a human who does not want to learn the screen. The screen, where it exists, is a fallback. The agent, where it works, is the primary interface. The data model has to be clean enough that the agent does not lie. The audit trail has to be tight enough that the CFO can sign the books. The license has to be open enough that the buyer never has to worry about lock-in.

Those four constraints rule out almost every existing ERP. They also rule out almost every “AI accounting” startup I have looked at, because most of them are AI bolted onto QuickBooks, which means QuickBooks is still the bottleneck.

I do not know if ERPClaw wins. I know the category is going to exist. I have spent eighteen years across enterprise software, including six years inside the consulting machine that built the previous category, and I have a strong opinion about what the new one looks like. ERPClaw is that opinion, expressed as code.

Where to go from here

If you read this far and you want to actually try it, the easiest path is one command:

clawhub install erpclaw

That installs the core ERP. Then you talk to your assistant. Tell it what your business is. It will set up your chart of accounts, your fiscal year, your tax rates, and suggest the modules you actually need. The whole thing takes about three minutes.

If you want to read first, the docs are a good place to start. If you want to follow along, the repo is at github.com/avansaber/erpclaw. A star helps. So does an issue. So does a pull request.

If you want to know more about the company behind it, that is AvanSaber. If you want to know more about me personally, my LinkedIn is the easiest way to get there, and I read every message that does not start with “circling back.”

The accountant in Greensburg retired a few years ago. I never got to tell her that her patience that day, sitting through the explanation she did not need, became the seed of a thing she would have actually used. I think she would have liked it. I think she would have ignored most of it and gone straight to the AP screen and posted the invoice in eight seconds without asking anyone for permission.

That is the ERP I want to build. That is the ERP I am building.