Gloat is now Synq.Read why

synq

People AI, governed
by construction.

Every request Synq acts on runs the same deterministic funnel a person would — identity, access, rules, approval, audit — decided before it ever touches a record.

Acts as you/Row + field access/Fail-closed/Append-only audit

People data is the hardest
data to hand an AI.

So Synq doesn't ask the model to behave.

Salary, home address, performance, health. Give an agent a service account and it sees all of it, for everyone. Synq's answer is that the agent never gets its own keys — it runs on your permissions, through a chain of deterministic checks that decide what's allowed before a single row is read.

Every request runs the same funnel.

1

Identity

Synq calls as the person asking — their token, their permissions. No service account, no god-token.

2

Access

A row filter and per-field mask are computed before any record is read. You see only what you're allowed to.

3

Rules & approval

A change is checked against your rules — warn, deny, or require approval — then held through an ordered chain until the last sign-off.

4

Audit

Actor, action, record, time — every step lands on an append-only trail you can export.

Synq acts as you — never around you.

When Synq reads or writes a record, it calls as the person asking: your identity, your permissions, on every tool call. There is no service account and no back door. Synq cannot see a field or a row you can't — the same holds when it runs inside Claude.

What's Dana Reyes's current salary?

What the funnel enforces.

Decided before the query runs

Access isn't a filter bolted onto results — it's the shape of the query itself. A dedicated decision service computes a row filter and a per-field mask for every request, before a single record is read. Withheld fields come back tagged restricted.

People · Engineeringselect compensation
Decision applied before any row is read
PersonLocationComp
PNPriya Nadella
Berlin$218,000
MOMarcus Okafor
Austin Restricted
LWLin Wei
Berlin Restricted

Four access models — Self, Group, Hierarchy, and Attribute. Ask for one more field and the rows themselves can narrow.

Writes run the same gate a person would

Every change Synq proposes is evaluated against your rules and routed through your approvals before it commits. An ordered chain gathers each sign-off; one rejection vetoes the request, and a change that goes stale while waiting is rejected — never silently applied.

Change · comp for Dana Reyes
Requires approval
⋮

A change to compensation.salary needs HR and Finance before it commits.

Requested→HR→Finance→Applied
Approval· Salary · $241k → $268kDR Dana Reyesheld
HR · signed→ Finance · pending

An HCM that shapes itself to you

Synq is the HCM — there's no blank schema to configure and no migration. Its data model is dynamic: describe what you need to track and Synq builds and extends it on the spot, queryable with no redeploy. Already run an HCM? Synq can sit on top of it, syncing both ways.

Employeedynamic schema
Full nameWork emailManager →DepartmentLocationEmployment type+ Visa status
Added by asking — indexed and queryable, no redeploy.

Synq is your system of record — or sits on top of your existing HCM, syncing both ways.

Everything lands on the record

Every action — by people and by Synq alike — is written to an append-only audit trail: the actor, the action, the record, the moment. The access decision is explainable from the same pass that enforced it, and you can export the whole thing to the warehouse you already use.

Activityappend-only · exportable
ActorActionRecord
SynqSynq (agent)
approvedcomp-change · Dana Reyes
PRP. Raman
updatedjob-title · Lin Wei
SynqSynq (agent)
requestedapproval · HR + Finance

Fail-closed, by design.

The hard part of access control is the edges — the over-reach, the probe, the outage. Synq's are defined, not incidental.

Over-reach

Empty, never widened

A query beyond your access

0 results
The filter resolved to nothing — access never widens on failure
Missing grant
No grant, identity, or field → empty, not “all rows”
Show comp for everyone in Finance
Probe

No existence leaks

Asking whether a record even exists

Indistinguishable
A denied row looks exactly like one that isn't there
No oracle
The shape of what you're refused never reveals what exists
Is there a comp record for the CEO?
Outage

It stops, cleanly

When the decision service can't answer

503 — request stops
It never falls back to running unfiltered
Closed by default
Closed is the only safe failure mode
(access-layer unavailable)

Built to be trusted — and independently proven.

Regional routing

LLM calls are routed within the regions your data-residency rules allow — set per company.

Model-agnostic

Azure OpenAI and AWS Bedrock behind one interface, swappable per task. No single-model lock-in.

Isolated per tenant

Every tenant's data lives in its own database schema, enforced on every query and every stream message.

Audited
SOC 2 Type II

Independently audited controls

Certified
ISO 27001

Information-security management

Compliant
GDPR

EU data protection

Questions about governance

Only the ones the person talking to it can. Synq calls every HR service as you, carrying your identity and permissions — so compensation, home address, and other restricted fields are masked for Synq exactly as they are for you. There is no service account that sees everything.

The access decision is computed before any data is read, so the row simply isn't in the result — and a denied row is indistinguishable from one that doesn't exist. You can't use Synq to probe for records you're not permitted to see.

No. Synq's people data is exposed to Claude over an MCP server that forwards the user's own identity, so the same access decision applies. Through Claude, each person still sees only what they'd see in Synq.

No. A compensation change is evaluated against your business rules, which require approval, and routed through an ordered approval chain. The write is held and only applied once every approver signs off — and if it goes stale while waiting, it's rejected rather than applied.

It's an append-only audit trail of actions — actor, action, record, time — that you can export to your own data warehouse. We don't claim a cryptographically sealed ledger; we keep an honest, exportable record and let your warehouse be the system of evidence.

Each tenant's data is isolated in its own database schema, and LLM calls are routed within the regions your residency rules allow. Synq is model-agnostic across Azure OpenAI and AWS Bedrock, so there's no single-provider lock-in.