Book time in our lab the way a manufacturer books time at a contract plant. Senior, US-based engineers design, build, test and run AI software inside the systems you already use. You own the product and the code.
2 benches open this quarter
BAY 01 · INTAKE
Discovery is paid, and it comes first. We sit with whoever does the work today, read your actual documents and look at your data. Then we write down what we’ll build, the examples it has to get right, and a price we stand behind.
BAY 02 · DEVELOPMENT
Getting a model to read a scanned deed is the easy part. The work is everything around it: matching names against your title records, posting bills to QuickBooks without doubling them, and handling the one vendor whose invoices arrive as phone photos. That’s where we spend our time.
Typical systems on the bench
BAY 03 · QUALITY CONTROL
Before launch we collect a few hundred real examples from you, including the messy ones your team complains about. Every version of the build runs against all of them. If a change makes even one of them worse, it doesn’t ship, and that rule keeps running after launch.
Sample eval run312 cases309 pass3 sent to a person
BAY 04 · SHIPPING
It runs in your environment, with documentation written for the people who’ll maintain it. If something we built breaks in the first 30 days, we fix it at no charge. After that, keep us on retainer or take it in-house; either works.
Example builds
Examples of what the lab can build across our core competencies. Client work stays confidential, so each one runs on sample data. Open any case to try it.
Capacity · Q4 2026Updated Oct 4, 2026
A Lab retainer holds engineering hours for you every month. Pick a size to see where you’d sit on the floor. Booked bays are dimmed.
One bench, held for you every month. Enough for one product in steady development, or several smaller tools.
Bench available this quarter: B-03.
Other ways to work
SPEC-201 · COMMISSIONED BUILDREV B
For one clearly defined product. Paid discovery produces the spec and a fixed price; everything after that is built and tested against that spec.
SPEC-301 · LAB HOURSREV B
For when you want an experienced engineer to look at a problem before you spend money on it. Sometimes the answer is “don’t build this yet,” and we’ll say so.
The monthly Lab Report
At the end of each month, retainer clients get a short report: what shipped, how the hours were spent, and whether the numbers we agreed to track moved. It’s the report we’d want if we were the ones paying.
BENCHSCALE · LAB REPORT
September 2026
01 Shipped
02 Hours · 80 of 80
03 Measured results
04 Next month
Signed off · Lead engineer
End of report
Sample report. Figures are illustrative.
Security
Carrollton, Texas
We work out of Carrollton, a short drive from most of Dallas–Fort Worth. Discovery happens at your office, at the desk of whoever does the job today, because that’s where the real process lives. The people you meet are the people who build it.
Questions
How retainers, builds, code ownership and your data work. Anything else, ask us directly.
Benchscale is a contract AI software lab in Dallas–Fort Worth. Operating companies reserve engineering capacity, and a senior, US-based team designs, builds, tests and runs AI software inside the systems they already use: databases, ERPs, accounting and document archives. Clients own the product and the code.
Benchscale builds AI that works with a company's own systems: document AI that turns invoices, deeds and contracts into structured records; assistants that answer questions from your ERP, databases and documents; MCP connectors that give ChatGPT, Claude or Copilot secure access to your systems; governed AI workflows; hardening for AI-built prototypes; and private AI models that run in your own environment.
A Lab retainer reserves 40, 80 or 120 engineering hours a month (Bench, Suite or Dedicated Lab), billed monthly in advance. Retainer clients get priority scheduling and a monthly Lab Report showing what shipped, how the hours were spent and what changed in the numbers you track.
A commissioned build is a fixed-scope project for one clearly defined product. It starts with paid discovery, which produces a written spec, test cases from your own examples and a fixed price. Development, testing and a documented delivery follow, with a 30-day warranty on the delivered build.
Yes. You own the product and the source code. It is deployed in your environment with documentation written for the people who will maintain it.
Every build ships with an eval suite built from your own examples, including the messy edge cases. Each version runs against it before launch, and any change that makes a passing case fail does not ship. The suite keeps running on every change after launch.
Your data stays in your environment wherever possible: your cloud account, your database, your network. When data has to move, it goes to US-hosted infrastructure you approve. AI models are used under commercial terms that don't allow training on your data, access is read-only by default, and access is logged and removed when the work ends. For data that can't leave at all, models can run privately in your own environment.
Yes, through MCP connectors. Benchscale builds an MCP server for each system, with sign-in through your existing single sign-on (Microsoft Entra ID, Okta or Google Workspace), role-based permissions from your directory groups, and an audit log of every request.
Benchscale starts in two industries where the team already has production experience: oil and gas land and title, and building-materials distribution. The same work (document AI, data connections and governed workflows) applies to most operating companies.
Benchscale is at 1536 Hutton Dr, Suite 100, Carrollton, Texas, in Dallas–Fort Worth. Discovery happens in person at your office across DFW, and the people you meet are the people who build it.
Submit a work order on the site, email hello@benchscalelabs.com or call (972) 571-3364. A lead engineer reads every work order and replies with a few questions and a time to talk, then proposes discovery, a retainer or lab hours.
Get in touch
Tell us what the lab should build. A lead engineer reads every work order and replies personally, usually with a few questions and a time to talk.
Rather email or call? hello@benchscalelabs.com · (972) 571-3364