AI on client documents, inside your own environment.
Soxoa sets up one workflow on a model that runs in an environment you control, your own hardware or your own cloud account, tests it on approved examples of your work, and writes down exactly what data goes where.
It fits firms whose documents cannot go into a public tool.
It fits if
- Staff are told not to paste client documents into public AI tools, and the work still gets done by hand.
- The work is documents: contracts, returns, claims, files, letters.
- Someone in the firm can say what data may be used, by whom, and where it may sit.
- You have a machine or a cloud account under your own control, or are willing to set one up.
It does not fit if
- An approved hosted tool already covers the work and your data rules allow it. That is usually the cheaper answer, and John will say so.
- You need a formal certification or a legal opinion. Soxoa provides neither.
- No example documents can be used, even inside your own environment.
How the three weeks run.
Write the data boundary first
Before any tool is chosen, we write down what data the workflow touches, where it may sit, who may reach it and what is kept.
Choose where it runs
A machine in your office, your own server, or your own cloud account are compared on quality, cost and upkeep. You decide.
Test models on approved examples
Candidate models are tried on your own approved documents. What each gets right and wrong is recorded.
Build the workflow with a review step
The workflow is built inside the boundary, and a named person reviews anything that needs judgment.
Hand it over
Documentation, upkeep notes and a named owner. If no option clears the bar, you get a written stop recommendation instead of a weak build.
What you have at the end.
| Deliverable | What it is |
|---|---|
| The workflow in operation | Running in your environment on real work, tested against the bar agreed at the start. |
| The data boundary | A written account of what leaves the environment, who can reach it and what is kept, for your own reviewers. |
| The evaluation record | Which options were tested on your examples and how each did, so the choice can be checked later. |
| Upkeep notes | Updates, licenses, what to watch, and who owns it on your side. |
One price, agreed before work starts.
$12,500Fixed price, one workflow
One price for one workflow, agreed in writing before work starts. Hardware and cloud charges are yours, and they are estimated before the build begins.
Included
- A written data boundary: what leaves the environment, who can reach it and what is kept
- The model and the workflow set up in an environment you control, tested on approved examples
- A human review step, documentation, and handoff to a named owner on your side
Not included
- Hardware and cloud charges, which are estimated before the build begins
- Software and model licenses where they apply
- Legal, security or regulatory sign-off, which stays with your firm
Boundaries.
- Outcomes
No results are promised. A model you run yourself can be weaker than a hosted one on some tasks, and the evaluation shows where before you commit.
- Private is a boundary
Running a model in your own environment does not by itself make a whole workflow private. The written boundary says what is inside it and what is not.
- Sign-off
Legal, security and regulatory sign-off stays with your firm. Soxoa gives no certification and no legal advice.
- A stop is a valid result
If nothing clears the bar, the pilot ends with a written stop recommendation. The price is for the work, whichever way it ends.
Questions people ask first.
Do we need to buy a server?
Not necessarily. A capable desktop, an existing server or your own cloud account are compared in the second step, with costs estimated before anything is bought.
Is a model we run ourselves as good as the public tools?
For some work yes, for some no. The pilot tests on your own documents first, so you see the difference before you rely on it.
Does this satisfy our regulator or our clients?
That is for your own reviewers to decide. The pilot gives them a written account of where the data goes, which is usually what they ask for first.
Who looks after it afterwards?
A named owner on your side, with the upkeep notes. Ongoing support is agreed separately if you want it.
Bring the work you cannot put in a public tool.
Thirty minutes with John. Describe the documents and the rule that keeps them out of public AI tools, and you will hear whether a private setup is worth it or a hosted tool would do.