AIEAdaptive Intelligence Engine
The memory layer that can prove what it remembers, why, and that it forgot.
In development · KINIX pilot first
- DCS products, starting with KINIX
- Hindi & Hinglish users
- Developers, after the pilot
- Languages it is built for: Hindi, Hinglish and English
- 3
- Ignore, ask or apply, by how confident it is
- 3 tiers
- For every change to what it remembers
- 1 receipt
- Launch journeys that must pass live before public launch
- 6
The problem isn’t too little memory. It’s wrong memory.
AI assistants that remember are no longer rare. What is rare is memory you can trust: one that keeps a preference to the right project, does not act on a guess, takes a correction the first time, and really forgets when asked.
AIE, the Adaptive Intelligence Engine, is a memory and confidence layer for DCS products. It learns a person’s preferences from conversation, applies them only when it is confident, lets the person correct or delete them, and gives a receipt for every change, in Hindi, Hinglish and English.
The memory layer that can prove what it remembers, why, and that it forgot.
Learn, scope, ask, correct, forget.
- 01Learn
From conversation
Preferences are picked up as you work, with the source they came from.
- 02Scope
Kept in the right place
Session, then project, then global, so rules for one project don’t leak into another.
- 03Ask
Confident, or it asks
Below the bar it is ignored; in the middle it is shown and asked about; only above it is applied.
- 04Correct
Fix it once
A correction becomes a rule that stops the same mistake coming back.
- 05Forget
Gone, and proven gone
Deletion reaches every copy, and the change is receipted.
It acts only when it is sure.
Every remembered preference carries a confidence. The thresholds are fixed in the engine, not a setting.
- Below 0.70
IgnoredNot used, not shown- 0.70 – 0.89
Shown and askedYou confirm before it counts- 0.90 and above
AppliedUsed, and you can see that it was
What is in development, and what waits for the pilot.
Nothing below is claimed as live yet. A capability counts only when a launch journey passes on the live system, certified by someone other than its author.
Scoped memory
Session, project and global, kept apart
In developmentConfidence tiers
Ignore, ask or apply
In developmentCorrections
Rules that stop a repeated mistake
In developmentConsent & private mode
Checked on every server write
In developmentVerified forget
Deletion across every copy, receipted
In developmentTime machine & export
What it knew when, with proof
In developmentBring your own keys
After a security test and a customer ask
PlannedConnectors & plugins
After the pilot shows the need
Planned
What AIE is built to refuse to remember.
Other people’s preferences
“My boss likes short emails” is not your preference, in English or in Hinglish.
Temporary instructions
A one-time request stays in the conversation and never becomes a standing rule.
Its own guesses
What the assistant assumed is not stored as something you said.
Injected instructions
Text that tries to plant a memory from a document or a page is refused.
Sensitive categories
Caste, religion, salary, immigration and health are blocked, in three scripts.
Anything it can’t classify
If it can’t tell what a fact is, it doesn’t keep it.
DCS products first, then developers.
KINIX
The first pilot runs on one KINIX workflow, with a controlled group of users.
People who work in Hindi and Hinglish
Memory that understands how India actually types, and asks when it is unsure.
Developers
A public quickstart and SDK follow the pilot result.
Where AIE stands today.
From AIE roadmap v3, checked 20 Oct 2026.
- Built in source
The memory, consent, correction and receipt work is built on a development branch and passes its tests.
- Next
Staging, then six launch journeys run on the live system and certified by QA. Nothing counts as live before that.
- Then
An independent security test and a fresh evaluation in Hindi, Hinglish and English, published with its method.
- Pilot
One bounded KINIX pilot. Pricing only after the pilot’s cost data.
AIE makes no performance claims until its evaluation is published. The workspace at aie.dcsai.ai is an earlier internal build.
Inside AIE
Pick a feature to see its screen, what it does, and how it works once you connect your own accounts and data. All values are demo data.
My AI Profile · scoped memory
SessionUse short bullet replies today
Project AtlasDates as DD-MM-YYYY
GlobalReply in Hinglish when I write Hinglish
Leak checkProject Atlas rules not applied to Project Orbit
Statusin development
Scoped memory
Session, project and global rules kept apart.
When it opens, AIE sits inside a DCS product, starting with one KINIX pilot workflow. Preferences picked up in conversation are filed as session, project or global, so a rule for one project never leaks into another.
Part of one ecosystem
Start with AIE.
Request early access, or talk to us about enterprise and sovereign deployment.