Security & Trust

Security at Matriq

Your raw data never leaves your warehouse. Nothing you give us is used to train any model. Here's the architecture, the controls, and the commitments behind those statements — and an honest account of where we are as a young company.

Pillar 01

Your data stays yours

Matriq queries your database directly via a read-only connector. We never copy, replicate, or store your raw business data on our servers — your tables stay in your warehouse, your results stay in your browser session.

Pillar 02

Least privilege by default

Read-only database connectors, scoped credentials, row-level filters, and full audit logs of every query the agent runs — so you always know what was asked and what was returned.

Pillar 03

No training on your data

Your queries, schemas, and results are never used to train foundation models. We use zero-retention LLM endpoints for inference and isolate every customer's memory layer.

Architecture

How data flows through Matriq

A read-only path from your warehouse to your team — with the agent's reasoning living in a separate, isolated memory layer.

Your Database Postgres · Snowflake · BigQuery · MySQL · Redshift
→
Matriq Agent Read-only queries · Schema introspection · Audit logged
→
Your Team Charts · Reports · Answers in plain English
What lives where. Your raw data never leaves your infrastructure. Matriq's memory layer (learned definitions, business context, prior corrections) is stored separately in an encrypted, customer-isolated workspace — and is fully exportable and deletable on request.
Data Handling

What we store — and what we don't

We store (encrypted, customer-isolated)
  • Connection metadata (host, database name, encrypted credentials)
  • Schema introspection: table names, column names, data types
  • Learned business definitions (e.g., "active user" = X)
  • Query results — cached so we can serve follow-up questions and rebuild reports for your own users. Stored in your isolated workspace; Matriq staff don't access it.
  • Query history and audit logs for your workspace
  • User accounts, workspace settings, and permissions
We never do
  • Copy or replicate raw rows out of your warehouse
  • Access your stored results as Matriq staff (the cache exists to serve your users, not us)
  • Mix one customer's data with another's — every workspace is isolated
  • Send data to any third party except the sub-processors listed below
  • Use anything you give us to train a foundation model — ours or our LLM providers'
Deployment

How Matriq is deployed today

We're being upfront about where we are today. Every Matriq workload currently runs on Matriq Cloud. Custom deployment for regulated industries is being scoped right now with our first design partners.

Available today

Matriq Cloud

Fully managed by Matriq on AWS. Connect your database in minutes — we host, patch, and scale the stack so your team doesn't have to. This is how every Matriq customer runs today.

  • Architecture aligned with SOC 2 Trust Services Criteria
  • Encryption in transit (TLS 1.3) and at rest (AES-256)
  • Logical isolation per customer workspace
  • Hosted in US-East (EU region on roadmap)
  • Zero-retention LLM endpoints (Anthropic & OpenAI)
In scoping with first design partners

Custom deployment

For healthcare, finance, public sector, and other regulated teams who need Matriq to run inside their own AWS, GCP, or Azure account. We're actively scoping this with our first design partners today. If you'd be one of them, we want to talk.

  • Runs inside your cloud account or your infrastructure
  • Customer-managed keys and IAM
  • Bring-your-own LLM where required
  • Built collaboratively with your security and platform teams
  • Email security@matriq.ai
Why we're being upfront about this. We've seen too many young vendors paste a "VPC, self-hosted, air-gapped" list on their security page and then quietly stall when an enterprise buyer asks for it. We'd rather tell you exactly where we are: Matriq Cloud is production-ready today, and custom deployment is real but currently scoped to design-partner engagements where we work closely with your security and platform teams. If that's the kind of partnership you're looking for, we want to talk.
Encryption & Infrastructure

Security at every layer

Encryption in transit

All connections use TLS 1.3. Database connections use SSL/TLS with certificate pinning where supported.

Live

Encryption at rest

AES-256 for all stored data including connection credentials, schema metadata, and learned definitions.

Live

Secrets management

Database credentials are encrypted with envelope encryption using AWS KMS. Decryption is scoped to your workspace's runtime only.

Live

Network isolation

Production workloads run in private subnets. No direct internet access for data planes. IP allowlisting available for warehouse connections.

Live

Customer-managed keys (BYOK)

Bring your own KMS keys for at-rest encryption of the memory layer. Available now for design-partner engagements running Matriq inside their own cloud account; Matriq Cloud BYOK on roadmap.

Roadmap

Backups & disaster recovery

Managed PostgreSQL on AWS RDS with automated daily backups and multi-AZ failover for the control plane. Formal RPO/RTO targets will be published with our first customer DPA.

Live
Authentication & Access

Who can do what — and how we know

SSO via Google & Microsoft

Out-of-the-box SSO for Google Workspace and Microsoft 365 accounts.

Live

SAML 2.0 / SCIM

Okta, Azure AD, OneLogin, and any SAML 2.0 identity provider. SCIM 2.0 for automated provisioning and de-provisioning.

Roadmap

Role-based access control

Workspace roles (Owner, Admin, Editor, Viewer) and per-data-source permissions. Limit who can connect databases, edit definitions, or publish reports.

Live

Read-only DB connectors

Matriq strongly recommends — and defaults to — read-only credentials for every connected warehouse. We will refuse to execute write statements.

Live

Full audit logs

Every prompt, every generated query, every result returned — logged with user, timestamp, and data source. Exportable to your SIEM.

Live

Row-level security passthrough

If your warehouse enforces row-level security on the connected role, Matriq inherits it automatically. The agent can only see what the credentials can see.

Live
Compliance

Where we stand on certifications

We'd rather be straight with you than paste a list of badges we haven't earned.

Matriq does not yet hold formal security certifications. Our architecture and controls are designed to align with the SOC 2 Trust Services Criteria from day one — read-only data paths, encryption in transit and at rest, audit logging on every query, isolated customer workspaces, zero-retention LLM endpoints, and a single point of accountability (the founder) for any incident.

We'll begin formal SOC 2 Type I work as we onboard our first paying customers. We'd rather earn the audit on real customer workloads than buy a certificate before we have any. Our approach to GDPR and CCPA is the same: we build to the principles today (data minimization, purpose limitation, deletion on request) and we'll publish a formal DPA at first customer onboarding.

Need to evaluate us right now? Email security@matriq.ai and we'll turn around a security questionnaire (CAIQ-Lite, SIG, or your own format) quickly. We're also happy to walk your security team through the architecture on a call.

Infrastructure & Vendors

What Matriq is built on

A small, deliberate set of infrastructure providers powers Matriq. When we onboard our first customers, we'll publish a formal sub-processor list and DPA covering this set.

Sub-processor Purpose Data accessed Region
Amazon Web Services Cloud infrastructure & hosting All workspace data (encrypted) US-East (us-east-1)
Anthropic LLM inference (zero-retention API) Schema metadata & natural language prompts only United States
OpenAI LLM inference (zero-retention API, optional) Schema metadata & natural language prompts only United States
Stripe Billing & payment processing Billing contact information only United States
Postmark Transactional email Email addresses & notification content United States
Cloudflare CDN, DDoS protection, WAF HTTP request metadata Global edge

Questions about any vendor on this list? Email security@matriq.ai.

Incident Response

If something goes wrong

Our commitments
  • The founder is on-call. Critical alerts page me directly.
  • Customer notification within 24 hours of any confirmed security incident affecting their data — we'll formalize this in our DPA at first customer onboarding
  • Post-incident review shared with affected customers within 5 business days
  • Written root-cause analysis for any P0/P1 incident
  • A public status page will go live alongside general availability
Reporting a vulnerability

If you've found a security issue, please tell us. We won't take legal action against good-faith research.

  • Email: security@matriq.ai
  • PGP key available on request
  • We'll acknowledge within 1 business day
  • We'll triage and respond with a remediation plan within 5 business days
  • Researchers credited publicly with permission
A note from the founder

Our commitments to you

Your raw data never leaves your warehouse. Matriq queries it directly via a read-only connector and returns answers to your team. We don't replicate your tables, and Matriq staff never look at your data.

Nothing you give us is ever used to train any model — ours or our LLM providers'. We use zero-retention API endpoints from Anthropic and OpenAI. Your queries, schemas, and learned business context stay isolated to your workspace.

If you want to read the source of the agent that touches your warehouse, we'll show it to you under NDA. And if you need a custom deployment inside your own cloud account, we're scoping that with our first design partners right now — we'd love to talk.

If you ever find a gap between what this page says and what we do, tell me directly. My email is below.

MT
Mohit Telang
Founder, Matriq · mohit@matriq.ai
Security FAQ

Questions security teams ask us

Your raw business data stays in your database. Matriq queries it directly via a read-only connector and returns answers to your team — we never copy or replicate your tables. The only things we store on our side are connection metadata (encrypted credentials, host, database name), schema introspection (table and column names), and the learned business context the agent builds over time. All of that is encrypted at rest and isolated per customer workspace.
Today, every Matriq workload runs on Matriq Cloud — fully managed by us on AWS with architecture aligned to SOC 2 Trust Services Criteria, encryption in transit and at rest, and zero-retention LLM endpoints. Custom deployment for regulated industries (running Matriq inside your own AWS, GCP, or Azure account) is being scoped right now with our first design partners. Email security@matriq.ai if you need a deployment option beyond Matriq Cloud and we'll talk through what we can offer right now.
Not yet, and we're being deliberately straight about that. Matriq's architecture and controls are designed to align with SOC 2 Trust Services Criteria from day one, but we have not engaged an auditor or completed a Type I attestation. We'll begin formal SOC 2 work as we onboard our first paying customers — we'd rather earn the audit on real customer workloads than buy a certificate before we have any. If you need a security questionnaire (CAIQ-Lite, SIG, or your custom format) to evaluate us today, email security@matriq.ai and we'll turn it around quickly.
SSO via Google Workspace and Microsoft 365 is available today. SAML 2.0 and SCIM provisioning (Okta, Azure AD, OneLogin) are on the roadmap, prioritized by customer demand. Inside Matriq, we offer role-based access control with workspace roles (Owner, Admin, Editor, Viewer) and per-data-source permissions. If your warehouse enforces row-level security on the connected role, Matriq inherits those restrictions automatically — the agent can only see what the credentials can see.
Matriq is built on a small, deliberate set of infrastructure providers: AWS (hosting), Anthropic and OpenAI (LLM inference via zero-retention APIs), Stripe (billing), Postmark (transactional email), and Cloudflare (CDN / WAF). The full table with purpose, data accessed, and region is above on this page. When we onboard our first customers, we'll publish a formal sub-processor list and DPA covering this set.
You can export or delete your workspace data at any time from settings. On account termination, all customer data — including the learned memory layer, audit logs, and connection metadata — is permanently deleted within 30 days. Encrypted backups age out within an additional 30 days. We will provide written confirmation of deletion on request.
Email security@matriq.ai. We'll acknowledge within 1 business day and respond with a triage and remediation plan within 5. We won't take legal action against good-faith security research, and we maintain a hall of fame for responsibly disclosed issues. PGP key available on request.
No. Your queries, schemas, and results are never used to train any foundation model — ours or our LLM providers'. We use the zero-retention API endpoints from Anthropic and OpenAI (these are contractual settings on the provider side, not defaults), which means prompts are not logged or used for model training. Each customer's memory layer (learned definitions, business context) is isolated to their workspace and never crosses into anyone else's environment.
Get in touch

Need a deeper security review?

We're happy to walk your security team through the architecture, fill out questionnaires (CAIQ, SIG, custom), or scope a custom-deployment engagement. Our first independent penetration test is scheduled before general availability — we'll share scope and findings under NDA when complete.

Email security@matriq.ai →
For general questions: hello@matriq.ai
Book a demo