How it works
Plain steps. No magic.
TensorCode is the software your engineers use to build AI tools your team can check and teach. Here's what that looks like day to day, without the jargon.
-
Point it at your sources
Your engineers connect the material your team already relies on: logs, tickets, reports, emails, documents. Every piece keeps a label saying where it came from.
For exampleThe ops logs, the vendor inbox, and the incident-reports folder.
-
Ask it questions
It answers in plain language and shows which sources back each part of the answer. Before you see an answer, it's checked against those sources. If it can't be backed up, you're told it isn't sure.
For example“Why did Tuesday's orders fail?”
“The payment database refused connections.” Sources: Ops log #17, Vendor email.
-
Your team checks the work
People confirm what's right and fix what's wrong. Each fix is saved, along with who made it. When the facts change, say a new report comes in, the answer is revised and the old one stays on record.
For exampleDana notes: “It was the database, not the network.” Saved as lesson #128.
-
It learns, and you keep it
On a schedule you choose, the AI trains on the saved lessons and on what actually happened after it acted. The result is saved as files on your own machines. Keep every version, and bring back any of them.
For exampleFriday night: retrain on this week's lessons. Monday: the new version is ready, and last week's is still saved.
The people part
Built for the people who know the work.
Your experts already know when an answer is wrong. TensorCode turns that know-how into something the software keeps, instead of something that walks out the door at 5 pm.
Giving feedback takes no technical skill. Your engineers build the screens; your experts just say what's right, and it's saved.
/media/team-review.webp · 1920×1080
Warm, candid editorial photograph from slightly above: three colleagues of different ages and backgrounds around a light oak table in a bright, modern office, calmly reviewing printed pages and a laptop together. One person places a small teal sticky flag (#3fd1b8) on a page; a coral (#ff8e6b) pencil mark is visible on another. Soft natural window light from the right, gentle shadows, shallow depth of field focused on hands and pages; faces relaxed and engaged, partly out of frame. Palette: warm paper white (#f7f6f2), pale oak, deep ink charcoal (#12151a) clothing, muted teal and coral accents. Mood: careful, collaborative, trustworthy, unhurried. Shot on 35mm, natural color grading, subtle film grain. No legible text on pages or screens, no logos, no brand names. Aspect ratio 16:9, 1920x1080.
What stays in your control
Your data. Your model. Your rules.
Your data
It runs where you run it. Nothing goes to an outside AI service unless your engineers choose to connect one.
Your model
What it learns is saved as files you own. Copy it, back it up, move it, or go back to last month's version.
Your rules
You decide how sure it has to be before it answers, and what it says when it isn't sure.
Straight talk
What it won't do.
We'd rather you hear this from us now than find it out in a pilot.
It isn't magic
It learns from the examples and corrections you give it. A handful of examples gets you a handful of learning.
It's early
TensorCode is in early access. It's best for teams who want to pilot it with us, not yet a boxed product.
Checks can be wrong
Software double-checks every answer, and software makes mistakes. That's why the receipts are always shown.
Questions
Common questions.
How is this different from a chatbot?
A typical chatbot gives you an answer and asks you to trust it. Tools built with TensorCode show their sources, check themselves, say when they aren't sure, and learn from your team's corrections, all on software you own.
Do we need an account with an AI company?
No. TensorCode can run entirely on your own computers. If you already use an AI provider, your engineers can connect it. That's your choice, not a requirement.
Does our data leave our building?
Not unless you send it. The software doesn't phone home, and it works with no internet connection once it's set up.
What do our engineers need?
People who write Python or TypeScript. The documentation is written for them, with runnable examples.
How much does it cost?
Nothing. The software is free and open source under the MIT license, and it runs on computers you already have. Your costs are your engineers' time and the machines you choose to run it on.
Is it ready for production?
It's in early access. Start with a pilot on one team's real questions, and measure it against how that team works today. We publish our own results, including the ones where the answer was “not yet,” so you know what to expect.
Start with one question.
Pick one question your team answers every day and hand this page to an engineer. It's free, and it runs on your own machines from day one.