Trust & Security

How we protect your data, secure your agents, and keep your operations running reliably.

Infrastructure & Encryption

OpsAgents runs on Google Cloud Platform (Cloud Run, Firebase Hosting, Cloud SQL). All data in transit is encrypted with TLS 1.2+. Data at rest is encrypted using AES-256 on GCP-managed storage. We use Secret Manager for all credential storage — secrets never appear in code or logs.

SOSA™ Framework

Every OpsAgents deployment runs under the SOSA™ (Supervised, Orchestrated, Secured, Agents) governance framework — a security-first architecture for autonomous AI agents. SOSA enforces:

Supervised — agents act within human-defined guardrails and escalate on uncertainty.
Orchestrated — all agent actions are logged, sequenced, and reversible where possible.
Secured — per-workspace credential scoping, with a dedicated per-tenant GCP project available for clients requiring physical isolation.
Agents — every agent run is accountable: it produces a verifiable audit trail. Administrative actions on our cloud infrastructure are additionally written to a locked, tamper-evident audit log that cannot be shortened or deleted for its full retention period.

Read the SOSA™ whitepaper →

Data Handling

Each client workspace is logically isolated — access is scoped so that one workspace does not read another's data.
We do not use your operational data to train AI models.
Application and agent activity logs are retained for 30 days; administrative audit logs are retained for 400 days under a locked retention policy that cannot be shortened. Deletion on request.
Third-party integrations (Trello, Notion, WhatsApp, Shopify) use OAuth or token-scoped credentials stored per-workspace in Secret Manager.

AI Model & Privacy

OpsAgents uses Claude (Anthropic) via AWS Bedrock and Vertex AI as the underlying LLM. Because inference runs through these enterprise cloud platforms, your prompts and responses are not used to train foundation models. See Anthropic's Privacy Policy and our Privacy Policy for details.

Subprocessors

We rely on a small set of vetted infrastructure providers to run the platform. Each is bound by its own security and data-processing commitments:

Google Cloud Platform — infrastructure hosting, compute, and database (Cloud Run, Firebase Hosting, Cloud SQL, Secret Manager).
Amazon Web Services (Bedrock) — enterprise serving of Claude models.
Anthropic — Claude foundation models (via the Bedrock and Vertex AI enterprise platforms).

Per-workspace integrations you choose to connect (e.g. Trello, Notion, WhatsApp, Shopify) act only on the data you authorize, under token-scoped credentials. The full, current subprocessor list is available on request.

Certifications & Compliance Posture

✓ HTTPS / TLS 1.2+ ✓ GCP AES-256 encryption at rest ✓ SOSA™ Agent Governance ✓ Secret Manager credential isolation SOC 2 Type II — in planning ISO 27001 — in planning ISO 42001 (AI management systems) — in planning GDPR — data processing agreement available on request

We're a fast-moving startup — certifications are in progress, not yet issued. We'll update this page as each milestone lands.

Uptime & Reliability

Agent infrastructure is designed for high availability on auto-scaling Cloud Run — capacity follows demand, with no single point of failure in the compute layer.
Managed, redundant GCP services back the storage and database layers.
Incidents affecting customers are communicated promptly to all affected workspaces, with updates until resolution. Contractual notification commitments are set out in the data processing agreement.

Data Residency

Data residency is determined per product, not globally — our services run across multiple Google Cloud regions, and we will tell you exactly where a given product's data sits rather than make a blanket claim. Israeli region hosting (me-west1, Tel Aviv) is in production use today, and clients with a residency requirement can be placed on a dedicated per-tenant GCP project pinned to a specific region.

If residency is a procurement requirement, ask us for the region of the specific product you are evaluating and we will confirm it in writing.

Access Control

Admin access to cloud infrastructure requires multi-factor authentication.
Production environments use Workload Identity Federation, and exporting a long-lived service account key is structurally impossible, not merely discouraged: the organization policy iam.disableServiceAccountKeyCreation is enforced across our GCP organization, so the create-key API is refused.
IAM follows least-privilege; access is scoped per service.

Security Contact

Vulnerability Disclosure

Found a security issue? Email us directly — reports are acknowledged promptly, and we provide status updates until the issue is resolved.

Security Questionnaires

Procurement or compliance review? Completed CAIQ/SIG-style responses are available on request — and we'll work through your own questionnaire with you.

Both reach us directly at security@opsagents.agency, also published at /.well-known/security.txt.

Talk to us about your security requirements