Stop answering the same question twice.
Payment terms, who signs off, how the month closes. Someone asks in chat, someone who knows stops to answer, and next month it is asked again. Kaltra answers from what your company already knows, in the chat tool your people already use. When the person in charge corrects it once, everyone gets the correction.
Get in touchHow it worksOne correction, everyone's rule
A member asks and gets an answer with its source. When the person in charge says it is wrong, the rule changes for everyone from then on. Nobody files anything, nobody writes a wiki page.
Nothing stored in the Kaltra Engine
The Kaltra Engine reads what a request needs and keeps nothing: no memory, no documents, no transcript. Your facts, rules and documents stay in your company's own environment, readable in plain form and yours to take.
Not another model
Kaltra is the intelligence on top of your company's data, not a chat window around a model. The language work is done by a model your company contracts directly, on your own key, under your own terms.
The problem
The people who know are not a help desk.
Your company already knows the answer. A few people hold it, and they are interrupted for it all day. Write it down and it goes stale by the next quarter, because nobody files the correction. Kaltra keeps the current answer, says where it came from, and lets the right person change it in one message.
- AskAnyone, in plain language, from the phone or the laptop.
- CorrectA "no, it's net 15 now" from the right person becomes the rule. From anyone else, it waits for that person's yes.
- DigestOnce a week, what changed and what is waiting for a yes.
How it works
Your data where your company runs. The Kaltra Engine in one place. One signed line between them.
The outpost
- Memory: facts, rules, preferences store
- Documents your people ask about files
- Credentials for your company's tools secrets
- Every prompt and answer passes through here your logs
- A small open-source service source
requests
The Kaltra Engine
- Reads what a request needs stateless
- Decides what to answer, what to remember the loop
- Upgraded in one place for everyone one release
- Modules: documents today, more over time shared
- Holds no copy of anything on the left nothing
The outpost runs in your company's own cloud account. The model is called from there, on your own key, never from the engine. Revoke the engine's key from the outpost and the engine can no longer reach anything. That is the off switch, and your company holds it.
What stays where
Nothing of yours is stored in the Kaltra Engine.
| Item | Lives in | Who can read it |
|---|---|---|
| Memory | Your environment | Your people, through Kaltra. Your admins, in plain readable form, yours to take. |
| Documents | Your environment | The groups your admin names for each source, and no one else. |
| Credentials | Your environment | Only your outpost. The engine never sees a secret value. |
| Prompts and answers | Your environment | Within reach of your own logs. They pass through the engine while a request runs and are not kept. |
| The Kaltra Engine | Kaltra | No one, beyond the request in flight. It holds no copy of anything above. |
Why not the chat product you already pay for
Your team already pays for an AI assistant. Kaltra does not replace the model. It is the brain around it.
| A chat product | Kaltra | |
|---|---|---|
| What it is | A model with a chat window. | The brain on top of your company's data. Not a model: it uses one your company contracts directly. |
| A correction | Fixes one person's chat. | Becomes a rule for everyone, set by the right person. |
| The company's memory | Search over connected documents, plus memory per person or per project. No company-wide rulebook one person corrects for everyone. | One shared memory inside your environment, readable and yours to take, shaped by the person you choose. |
| The model | The vendor's own. | A model your company contracts directly, on its own key, under its own terms. |
The model sees the same content either way. The difference is what is kept, and where: your memory and your documents stay in your environment, and the Kaltra Engine keeps nothing.
"Nothing to defend, because nothing is held. Your company's memory lives in your company's environment, and the engine can only read it while your company allows it."
The principle the Kaltra Engine is built on.
Questions
The ones every company asks first.
- Is this another chat assistant?
- No. Kaltra is not a model and has no chat window of its own. It is the memory, the rules and the judgement on top of your company's data, reached from the chat tool your people already use. The language model is contracted by your company directly, on its own key.
- Which chat tool?
- One chat connection per company, attached when your company is set up. Which one is part of the first conversation.
- Who runs it inside our company?
- One designated person corrects and approves; the weekly digest is theirs. Nothing becomes a rule for everyone without that yes.
- What does our IT have to run?
- One small open-source service inside your company's own environment, plus the registrations any integration needs. New releases of that service are applied on your schedule, never pushed into your environment from outside. The engine is upgraded in one place, with nothing for you to redeploy.
- What if your company stops using Kaltra?
- Your company's memory is in its own environment, in plain readable form. Take it to any other tool; nothing of it was ever stored anywhere else.
- Does anything learned in our brain reach another company?
- No. Not the memory, not the associations, not anonymised. Only the engine and the generic modules are shared, and those hold no company data.
- How is it priced?
- A monthly fee for the engine, counted from the day your people start using it, not from the day the build begins. A fixed fee for each module your company turns on. Model spend on your company's own key, and whatever your own environment bills for running the outpost. A module your company needs that does not exist yet starts with a conversation.
Contact
Get in touch.
One address, read by the people who build Kaltra. Say what your company keeps asking itself.
A reply comes from a person, not a sequence.