Unscripted SaaS

Unscripted SaaS · Interview

Tom Girgash on Building Taproot, the Memory Layer for Solo AI Operators

with Tom Girgash — founder of Taproot, building full-time after quitting his job three weeks earlier

All Episodes

Why this matters for SaaS: Tom Girgash quit an ERP consulting job to build Taproot, a Mac-native memory layer that stores your knowledge in local files and pipes it into Claude, ChatGPT, Codex and Claude Code over MCP. It is a first-time-founder story at the earliest possible stage — about twenty signups — and half the conversation turns into a working consult on the three problems every early SaaS founder hits at once: terminology, pricing, and distribution.

The conversation

Episode overview

Jeremy Rivera talks with Tom Girgash, a mid-twenties Ohio State grad and current University of Cincinnati information-systems master’s student who sold bearings and hoses in the field, did a short stint as an ERP consultant at Cohen & Company, and then quit to build software full time. His product, Taproot, is a knowledge layer for heavy AI users: your files live on your own computer, and any tool that speaks MCP can pull that context in. The pitch rests on one observation — an LLM generates a fresh answer every time, so the reasoning it hands back is only worth anything when it is grounded in your actual work.

What we get into

Highlights

  • Why “past chats” is not memory — and why you can’t tell whether what surfaced is true
  • Local-first data ownership: the knowledge base is files on your machine, not a record inside someone else’s product
  • MCP as the cross-tooling layer — one base, connectable from Claude, Claude Code, ChatGPT, Codex
  • The mom test, and how Tom landed on “memory layer” after months of bad terminology
  • The origin story: a web-scraping cold-outreach tool, then an MVP built for a man named Tim
  • Pricing to a pain point — targeting the user about to jump from the $20 tier to $100–200, not the free tinkerer
  • Why distribution, not code, is the hard part — and what a bootstrapped founder can do without ad spend

The core idea

A memory layer grounded in your own work

The problem Tom is solving is his own. Heavy AI users have no place where their knowledge actually lives, so the reasoning they get back is untethered from what they have already done. Taproot lets you file and organize information quickly, synthesize across clients and projects, and keep everything on your own computer — so what you build inside Claude or ChatGPT stays yours. He frames the payoff bluntly: if the app surfaces the correct information it is useful, and if it just hallucinates it is not.

Anything that takes MCP can connect — Codex, ChatGPT, Claude, Claude Code. Tom calls that the beauty of it. As a self-described idea guy he will hash out a software plan in Claude in the browser, then bring that exact context into his coding environment, then pull his go-to-market notes and channel-specific content back out later, all grounded in things he wrote himself.

“I’m not trying to build something just for a random group of people that I really don’t know much about. It’s for solo operators, people like myself that are using AI a lot, dealing with different contexts.”

— Tom Girgash, Taproot

Messaging

The mom test, and the question that opens the door

Tom says terminology has been the single hardest part of the build. His mom and his sister did not understand what Taproot was, which is the only market research a first-time founder really needs at that stage. He converged on “memory layer” language and an opening question that lands every time: where does all the stuff you have done in Claude or ChatGPT actually live? People answer “in past chats” — and then admit they cannot find it, and cannot tell whether what surfaced is real. That gap is the whole product.

The first wave is about twenty signups after a couple of weeks, and it is Mac-only, because Tom is a Mac native and that was the fastest path to something real. The bigger version he originally imagined is a firm brain: every employee keeps their own layer, and the team taps a shared, cohesive context on top. That is later. For now the wedge is solo operators, a category he expects to grow considerably as one person keeps being able to do more.

The hard part

Distribution is the last moat

Asked what he has not solved, Tom does not hesitate: distribution and marketing at scale. He will not spend on ads yet, so he needs an organic engine. Forums and Reddit are real avenues and also a minefield — people do not like being sold to, and he has been burned trying. He also named a trap that is specific to this moment: with AI you can do so much that identifying the one next step that actually moves the needle gets harder, and it is very easy to spend three weeks tweaking a website when talking to ten users would have taught you more.

“I think cracking distribution and marketing has been, is going to be the toughest challenge and so far it is, you know, the toughest challenge to do at scale.”

— Tom Girgash, Taproot

His thesis about where that leads is the most interesting thing he said. As software gets commoditized, distribution becomes the most important component of building any business — and because audiences are exhausted by AI slop, the durable advantage swings back toward being recognizably a person.

“I think being human, being as human as possible, connecting with your audience on a very human level, it’s going to be something that’s going to be pretty large going forward.”

— Tom Girgash, Taproot

From the host chair

Pricing, churn, and paid feedback

I ran product for Raven Tools, and the lesson that transfers here is about the interaction between your pricing model and your churn. “They left because of price” is not actionable information until you separate competitor price, budget, and value-to-price mismatch — three completely different problems that produce the same exit survey answer.

Tom is, in effect, drafting on the wind of Claude subscriptions. His real buyer is not the free tinkerer; it is the user already running out of tokens on the $20–25 tier and looking at $100–200. A higher price demands higher delivered value, while a low-dollar add-on can squeak by on convenience — so the price point has to match the pain point he is actually relieving. My concrete suggestion was an AppSumo-style launch: trade a lifetime discount for a committed early cohort and you are close to getting paid for a feedback audience, which is exactly what he needs before he can set a real price.

Ideas worth stealing

Key takeaways

  • Build for a user you actually are. Tom’s ICP is himself — solo operators drowning in context — and it keeps the roadmap honest.
  • Run the mom test on your positioning before you run ads against it.
  • Local-first is a feature, not an implementation detail: the pitch is that the knowledge stays yours.
  • Ground the model or don’t bother. Reasoning is only useful when it sits on top of your real work.
  • Price to the pain point, and find the tier your buyer is already about to outgrow.
  • Ship for feedback, not polish — weeks of website tweaks will not teach you what ten user conversations will.

Connect with Tom

Where to find him

Try it: taproothq.com · Tom is on X as @TGIRGS and on LinkedIn as Thomas Girgash. Taproot is Mac-only at the time of recording, with a Windows build not yet on the board.

Stay in the loop

Get new SaaS founder interviews in your inbox

The sharpest takes on growth, product, and go-to-market — delivered as each new episode drops.

Subscribe Free →All Episodes