--- title: "Security — agent calls checked, logged, and revocable · Elva" description: "Elva secures agent access with five ordered checks on every tool call: identity, tool scope, PII redaction, rate limits, and a full audit log. Revocation is instant." canonical: "https://getelva.ai/security" lastModified: "2026-09-01T23:11:23.016Z" --- Security # Agents get a door, never a spare key Your upstream credentials never leave Elva. Every tool call is identified, scoped, redacted, throttled, and logged. This page is written for the person who has to sign off on that. [Request the security pack](mailto:hello@getelva.ai) [Enterprise options](https://getelva.ai/enterprise.md) 5 CHECKS BEFORE YOUR API 0 UNSCOPED TOOLS ALLOWED Every CALL LOGGED, NO SAMPLING Seconds TO REVOKE A KEY 01 Request path ## Five checks, in order, fail closed Nothing reaches your API until identity, scope, redaction, rate, and audit have all passed. A failure at any step denies the call and records why. There is no bypass path and no unscoped mode. EVALUATING A TOOL CALL get\_payment\_status · Claude Desktop 01 Identity OAuth2 token or managed key, resolved to a named agent pass 02 Tool scope Is this key allowed to call this specific tool pass 03 Field redaction PII and internal fields stripped per contract pass 04 Rate and quota Per key and per tool, retry storms contained pass 05 Audit record Actor, tool, arguments hash, latency, outcome pass DECISION LOG ✓ identity: agent claude-desktop, key ...8f2a ✓ scope: read:payments present ✓ redaction: customer\_email removed ✓ rate: 41 of 600 per minute ✓ audit: written, 98ms, allowed 02 Data handling ## What we store, and what we never see Elva reads your code to derive specs; it does not need your production data. Request and response bodies are not retained by default, and PII fields are redacted at the gateway before they reach an agent. Source code access is read-only and scoped A least-privilege app installation on the repos you choose. We read routes, handlers, and types. No write access, no other repos. Bodies are not retained by default Logs record the tool, actor, arguments hash, latency, and outcome. Full payload capture is opt-in, per contract, with its own retention. Encrypted in transit and at rest TLS 1.3 everywhere, AES-256 at rest, per-tenant isolation. Private cloud installs can use your own KMS keys. Your credentials stay in the gateway Upstream API secrets live in Elva’s vault, never in an agent config or a partner machine. Rotate without redistributing anything. 03 Posture ## Compliance, and how we work toward it We would rather tell you our real status than imply a badge we do not hold. Ask us for the current pack and we will send it with dates attached. CONTROL STATUS NOTES SOC 2 Type II in progress Audit underway, report available under NDA when complete GDPR compliant DPA available, EU data residency on request Encryption TLS 1.3 / AES-256 In transit and at rest, per-tenant isolation SSO / SAML and SCIM available Enterprise plans, your IdP owns publish rights Penetration testing annual Third-party, summary shared under NDA Vulnerability disclosure open security@theneo.io, one business day response Data residency configurable Regional hosting, private cloud, or on-prem 04 When something goes wrong ## Revoke in seconds, explain in minutes One click kills a key, an agent, or a whole server, and the log keeps everything it did up to that moment. That is the difference between an incident and an unanswerable question. Instant revocation Keys, agents, and servers can be killed independently. In-flight calls are cut, not drained. Anomaly alerts Error-rate spikes, scope escalation attempts, and retry storms page you before a partner notices. Forensics on demand Filter the full log by agent, key, tool, or outcome, and export to your SIEM as JSON, CSV, or webhook. Report something to [security@theneo.io](mailto:security@theneo.io). We respond within one business day and credit every valid finding. ## FAQ ### How does Elva secure agent calls? Five ordered checks on every tool call: identity (OAuth2 token or managed key resolved to a named agent), tool scope, field redaction per contract, rate and quota, and an audit record. ### Does Elva expose my API credentials to agents? No. Upstream credentials stay in the gateway. Agents hold scoped keys that can be limited per tool and revoked instantly, one key per consumer. ### What ends up in the audit log? Every call records the actor, the tool, an arguments hash, latency, and the outcome, queryable per key and per tool. ### How do I report a security issue? Email security@theneo.io. Reports get a response within one business day. RELATED READING [ Governance Aug 29 · 8 min How teams actually control what AI agents can access Read → ](https://getelva.ai/blog/agent-permissions-and-data-access-governance.md)[ MCP Aug 30 · 9 min Tools to build governed MCP servers for agents, compared Read → ](https://getelva.ai/blog/governed-mcp-servers-for-enterprise-agents.md)[ Governance Aug 27 · 4 min An API governance checklist that fits in one pipeline Read → ](https://getelva.ai/blog/api-governance-checklist.md) ## Send this page to your security team They will ask about scopes, logs, residency, and revocation. All of it is answered here. [Request the security pack](mailto:hello@getelva.ai) [See plans](https://getelva.ai/pricing.md)