Six companies run on software we wrote by hand.
ERP, point of sale, dispatch and AI grounded in your own records. Built in St. Louis, running in production, and you can open every one of them below.
(586) 300-4975ScrollSophia is on your phone now. Three apps for Android.
An assistant that runs on the phone itself, a gallery that knows where every file came from, and a home screen that ties them together. Download them straight from us.
Sophia AssistantLaunching on AndroidVoice, reminders and everyday commands, with the thinking done on the phone.
Sophia VaultLaunching on AndroidPhotos, documents and AI media, filed, sorted and searchable.
Sophia LauncherLaunching on AndroidA home screen with Sophia built in, arranged the way you work.
- AndroidThe Sophia appsAll three apps, what each one does, and direct downloads.
- ServiceCustom ERP softwareA double-entry ledger, order to cash and procure to pay, on one system.
- ServiceMulti-store point of saleSeveral stores on one system, each store's stock kept its own.
- ServiceDispatch and field serviceFrom the call to the closed job, entered once.
- ResearchSophia XE5Our compositional language model: specialists composed on every token.
Open any of these. Somebody runs their business on it.
Not screenshots of launch-day sites. Systems in daily use, and the line under each one is what the system does rather than what the page looks like.
A double-entry ERP with a live wholesale market wired in. It matches what you are selling against who is buying and prices the deal from real market signal.
Each one states what we delivered, what it connects to, and what we have not verified.
Your business, live on one screen.
Six kinds of system, and every one is already running for a customer. The dashboard on the right is the shape of it: the office in a browser, the technician on a phone, one set of records underneath both.
-
One system that holds the whole operation, instead of five that disagree and a spreadsheet that settles it.
Two running -
Several stores on one system, with each store's stock kept its own.
Three stores, live -
Customer to call to schedule to technician to closed job, on one board.
Two service companies -
Integrations
Payments, accounting, supplier feeds. Whatever is already in the building stays.
Every build -
AI on your own records
Answers that cite the row they came from, so they can be checked.
Shipped in the EBS -
Staff and back office
Time clock, scheduling, permissions, in-store displays.
Retail and field service
- Revenue this month $184,200 12% up 78% of goal
- Jobs closed 312 8% up
- Receivables over 30 days $6,140 21% down
Not a customer's numbers. The shape of a service company's month, held in one system: dispatch, stock, invoices and the ledger they all post to.
The tooling we built to ship them.
A small team shipping whole business systems needs tooling nobody sells, so we wrote it. These run every build on this page, and several turned out to be worth handing to other people.
The problem each one exists for, the decision that settled it, and the numbers.
Every coding agent you own, in one window.
Most agent CLIs are one conversation in one terminal. Open a second one and it knows nothing about the first. This runs them all as a supervised set, each in its own git worktree, so two of them editing the same file cannot collide. You read the diffs, then you merge. Nothing merges on its own.
Three agents in one session, each in its own pane. The bar across the top is what every agent is doing right now.
Markdown to PDF, with no TeX installed.
texwright turns Markdown, LaTeX or a JSON document tree into PDF or HTML with its own TeX-style engine, and it needs only Node.js 20 or later. Its check command returns diagnostics, most with a fix an agent can apply, and an MCP server is included.
npm install -g https://sophiaxt.com/texwright/texwright-latest.tgz
Your staff can use AI. Your clients' data stays yours.
An AI assistant for law, accounting, insurance and other firms that hold client data. Before any message reaches the AI, Safe-Desk swaps Social Security numbers, account numbers, names and the rest for placeholders, then puts the real values back into the answer. Chat messages are never stored. Files uploaded for the AI connector, and the documents finished from them, are kept encrypted for 7 days, and encrypted backup copies are gone within 14 days after that. It also covers ChatGPT, Claude, Gemini, Copilot, Perplexity and DeepSeek through a browser extension, and your own software through an API.
Guides on desk.sophiaxt.com: AI for accounting and tax firms, AI for law firms, AI for insurance agencies, AI for financial advisors and is it safe to use ChatGPT with client data?
One message through the gate
Your staff type
her SSN 123-45-6789 is on file
The AI sees
her SSN [SSN_1] is on file
Your staff get back
her SSN 123-45-6789 is on file
The owner's log
Alerted: Social Security number
Agents that ship pull requests, not merges.
A board of tickets. Each ticket gets its own git checkout and its own agent. A separate verifier lane reads the result and can fail it, and a failed ticket cannot be merged. When a person overrides that, the override is recorded with their name on it.
It is invite-only, and not because of scarcity. Boards are not partitioned per account yet, so everyone signed in sees the same board. Until that is fixed we are not selling it, and we are not going to pretend otherwise.
The papers are part of the product.
We publish the method, the failures and the experiment log, not a summary of the good runs. Anchor token masking is written up with the numbers that made us keep it and the ones that nearly made us throw it out.
A lab that only shows you its wins is showing you a brochure.
\mathcal{L}_{\text{anchor}}=\mathbb{E}_{t\sim\mathcal{U}(0,1)}\;\mathbb{E}_{x_t\sim q(\,\cdot\mid x_0,\,t,\,m)}\!\left[\sum_{i\in A} w(t)\,\log p_\theta\!\left(x_0^{(i)}\mid x_t,\,x^{Q}\right)\right]The sum runs over answer positions only. Question tokens stay in the conditioning term on the right and never enter the corruption q(· | x₀, t, m). Flip that asymmetry and the advantage goes with it, which is how we know it is the asymmetry doing the work rather than the extra objective.
SOPHIA XT is the engineering arm of 3 Point Forward.
3 Point Forward
3 Point Forward does the business work: diagnosing what is actually wrong with an operation, planning the implementation, applied-AI consulting, training the staff who will run it, and managing the engagement.
3pointforward.comSophia XT
SOPHIA XT does the engineering: custom business software, integrations, AI engineering, developer tooling and the research underneath it.
One company, two jobs. 3 Point Forward defines the operational problem and the implementation plan; SOPHIA XT builds what the plan needs. A company that already knows what it wants built can go straight to SOPHIA XT, and a company that needs the problem defined first should start at 3 Point Forward. Either door reaches the same team.
Tell us what it has to do.
Not a specification. A few sentences about the work the software has to carry and what it has to talk to. We will tell you whether it is a thing we have built before, and say so plainly when it is not.
If the operational problem is not settled yet — which is more often the case than not — the consulting side runs a short business systems assessment that finds it first.
Business systems assessment at 3 Point Forward
Tell us what you want to build.
A free consultation with Thomas Garren. Scope gets written down before code, you get working software in weeks rather than at the end, and somebody still answers the phone a year later. You will also get a straight answer about whether we are the right team.
Or email contactus@sophiaxt.com