Architecture
The diagram below shows how data moves through AgentMark, the trust boundaries it crosses, and where it rests. Share it with your security team as-is; it lives in the same repository as the product and gets updated with every architecture change. Key boundaries:- PII masking runs inside your process. Redaction happens before telemetry leaves your infrastructure, so masked values never reach AgentMark.
- The gateway authenticates every request against per-app, permission-scoped API keys before storing anything.
- Telemetry and application data live in separate stores. ClickHouse holds traces and scores; Supabase Postgres holds accounts, prompts, and metadata behind row-level security.
- Third-party AI providers only see what a feature needs. The subprocessor list names each provider, its purpose, and the data involved.
Authentication
AgentMark supports email/password authentication for all tiers and SAML 2.0 SSO for Team and Enterprise tiers, covering Azure AD, Okta, Google Workspace, and any SAML 2.0-compliant IdP, with domain allowlisting, SSO enforcement, and attribute mapping. See SSO for configuration.Data protection
PII masking
Redact sensitive data from traces before it leaves your application. Masking runs in your application process, so it redacts configured attributes before the OTel exporter sends them.AGENTMARK_HIDE_INPUTS=true or AGENTMARK_HIDE_OUTPUTS=true to redact LLM request inputs or response outputs to [REDACTED] before export. The flags cover every content-bearing input and output attribute: the AgentMark SDK keys (such as agentmark.request.input and gen_ai.request.input), the Vercel AI SDK telemetry keys (ai.prompt*, ai.response*), and the OTel GenAI message and tool keys. Attributes that carry no message content, such as gen_ai.request.model, aren’t redacted. The full attribute list is in the masked attributes reference.
Full PII masking docs →
Encryption
- In transit: all API communication uses TLS 1.2+ (terminated at Cloudflare)
- At rest: data stored in Supabase (PostgreSQL) and ClickHouse with provider-managed encryption at rest
Provider API keys
AgentMark stores the AI provider keys you configure per environment (OPENAI_API_KEY, ANTHROPIC_API_KEY, etc.) as encrypted secrets in the AgentMark vault, using authenticated encryption with a root key held outside the application database.
- Scoped per app: one app’s keys are never visible to another
- Decrypted only when needed: AgentMark pulls values from the vault and provides them to the serving machine at runtime, or when an authorized Dashboard user explicitly clicks “reveal” on a single variable
- Never written to logs: request logs and trace exports omit env var values
- Deleted on demand: removing a variable from the Dashboard deletes the underlying vault secret in the same transaction
AgentMark API key security
AgentMark issues API keys (the keys your code uses to authenticate with AgentMark) with per-tier rate limiting, each scoped to a single app, so no single key grants access across your organization. Each key carries a permission set enforced by the gateway on every request. See API keys for the Dashboard walkthrough and Users and access control for role presets and the permission catalog.No training on your data
AgentMark never uses your data (prompts, traces, datasets, evaluation results, or other telemetry) to train AI or machine learning models. Features that rely on third-party AI providers, such as trace topics, run under commercial API terms that prohibit the provider from training on your data. The Terms and Privacy Policy state this commitment, and the subprocessor list names every AI provider involved.Subprocessors
AgentMark publishes a versioned list of every subprocessor that handles customer data, including what each one processes and where it runs, at agentmark.co/subprocessors. That page gets updated before a new subprocessor starts handling customer data.Data residency
If you have regional hosting, on-prem, or data residency requirements, reach out. Enterprise deals can accommodate custom arrangements.Data retention
Your tier’s retention window governs telemetry end to end: how far back the Dashboard and API can query, and when AgentMark physically deletes the underlying data (traces, spans, scores, and offloaded payloads) from its stores.
Telemetry older than the window no longer appears in query results, and a daily cleanup job hard-deletes it from storage. The cleanup job measures age from ingestion time, so late-arriving data still gets its full window. Enterprise agreements can tune retention per tenant through an entitlement override: longer or unlimited windows for audit needs, or shorter windows to meet data-minimization requirements. Reach out to configure it.
Temporary support access
When Enterprise customers need hands-on support, AgentMark support engineers can grant themselves temporary read-only access to a tenant’s data. Access auto-expires after 24 hours and requires customer permission confirmation. AgentMark records all access grants in an immutable audit log.Have questions?
Reach out any time:
- Email the team at hello@agentmark.co for support
- Schedule an Enterprise Demo to learn about AgentMark’s business solutions