Home / Company
Accountability was built for a world where people made the changes
Every control in your software development lifecycle assumes a named human at the other end: a reviewer, an approver, a leaver to offboard. That assumption held for thirty years. It stopped holding the week a coding assistant merged its first pull request under someone else's credential.
Quantlyx exists to make the assumption true again — not by slowing anything down, but by making it possible to say, with evidence, who or what actually made a change and whether anyone approved it.
What was approved
- Change tickets, release approvals and their approvers
- Code ownership, review requirements and exemptions
- Entitlements, certifications and revocations
- The accountable human, named on the record
What was done
- Commits, merges and who or what authored them
- Build identities, signing keys and artefact provenance
- Deployments, rollbacks and the principal behind each
- Runtime credentials and every checkout of them
An approval is a decision in one register and a machine identity in the other. Only the difference tells you who really shipped the change.
What we hold ourselves to
Every finding carries its evidence
No confidence scores standing in for proof. If Quantlyx says a bot deployed after its entitlement was revoked, it hands you the deployment record and the revocation record, both with identifiers you can look up yourself.
We publish what we do not cover
The threat mapping names the entries Quantlyx addresses in part and the ones it does nothing for, and points at the tooling that does.
The reconciliation core never takes write access
Any capability that acts on a finding is a separate, opt-in component with its own scoped, short-lived credential and a human approving each action. Discovery and response never share a permission.
A clean result is a valid result
If a pilot reconciles cleanly, we say so and you keep the report. We would rather lose the deal than manufacture a finding to justify one.
Where Quantlyx goes next
Two things follow reconciliation: asking the register questions in plain language, and acting on a finding without copying identifiers between four consoles at midnight. Both follow the same rule: the part that reads never gains the power to act.
The reconciliation core
- One register of every actor
- Differences with evidence
- Owner routing and closure tracking
An assistant that only knows your register
- Ask in plain language, get the actors and evidence
- Every answer cites the source record
- Says it does not know rather than guessing
QuantQFix guided response
- A playbook per finding type
- Actions proposed, never taken unattended
- Scoped, short-lived credentials per action
The reconciliation core is shipping and QuantQFix is ready for pilot. The assistant is in design and has no date.
An assistant distilled down to one job
A compact model distilled for this domain, deployed inside your tenant and pointed at nothing except the register it already built. No tools, no write path, no route to the internet.
Why distilled rather than large
- Runs on hardware you already have, inside the boundary, with no data egress
- A narrow domain can be measured against a fixed set of questions
- Cheap enough to run per query without a per-seat licence conversation
- No third-party model provider in your change record
What it will refuse to do
- Answer from anything it cannot cite
- Read your source code, tickets or documents
- Take an action, open a ticket or change a permission
- Speculate about intent, blame a person or infer a motive
Design intent, not a shipped feature. Wording and behaviour will change before release.
Response with a person on every action
A finding today ends with an owner and a piece of evidence. What happens next is mechanical and slow: open the privileged access platform, find the credential, rotate it, open the governance platform, close the entitlement, write the change record. That is where findings go to age out.
Guided response packages each step as a playbook the owner reviews and approves. Quantlyx proposes; a human decides; a scoped credential executes; the whole exchange lands in the evidence trail.
Finding
With evidence
Proposed action
A playbook step
Human approval
Cannot be skipped
Scoped execution
Short-lived, one use
Evidence trail
Proposal and outcome
Declined at the gate, the finding simply stays open. The approval gate is not configurable away. No playbook runs unattended.
See the playbooks for each threat
Separate credential, separate component
The reconciliation core keeps its read-only credential and never gains another. Response runs as its own component with its own permission, granted per playbook and withdrawable without losing your findings.
Short-lived and scoped to one action
A credential issued to rotate one secret can rotate that secret, once, within a window. No standing privilege sits waiting for the next incident.
The proposal is evidence too
What was proposed, who approved it, what actually ran and what the source system reported afterwards all land in the same trail as the finding that started it.
Reversible first
The early playbooks are the ones you can undo: rotating a secret, removing a role, opening a change record. Anything that could stop a release stays a proposal you carry out yourself.
QuantQFix is ready for pilot, not yet generally available. Prove it in a pilot on one lifecycle path before relying on it in a purchasing decision.
ID entitlement. Rank by risk. Remediate.
Start with a two-week, read-only pilot against one lifecycle path. If nothing turns up, you keep a clean report.