Revocable virtual keys
Give each application a broker-issued credential that can expire, suspend, or revoke without rotating a vendor master key.
USE CASE / AI AGENT APPLICATIONS
Give every agent one governed path to enterprise LLMs.
Replace direct vendor master keys with broker-issued virtual keys and apply model routing, budgets, RPM/TPM limits, guardrails, and audit to every agent call.
CONTROL OUTCOMES
Move from fragmented access to enforceable policy.
Scoped model access
Attach the key to an organization, team, or project and govern the model aliases and deployments it can use.
Budget and safety gates
Apply token or currency budgets, RPM/TPM limits, and request/response guardrail policy in one request path.
Call-level accountability
Record the key, provider, upstream model, tokens, spend, guardrail decisions, latency, and outcome.
REFERENCE ARCHITECTURE
Controls execute in the request path.
Mint an application key
Issue a virtual key for the agent application under its organization, team, or project scope.
Call one endpoint
Point the existing SDK at the on-prem OpenAI-compatible gateway and request a stable model alias.
Govern and route
Apply limits and guardrails, then load-balance or fail over across configured vendor deployments.
Meter and audit
Accrue scoped usage and write the complete provider and model call record to Postgres.
GATEWAY TELEMETRY
ILLUSTRATIVE REFERENCE VIEWTIMEVIRTUAL KEYMODEL ALIASPROVIDEROUTCOME
14:32:08vk_finance_07fastanthropicSUCCESS
14:31:55vk_support_02reasoningopenaiFALLBACK
14:31:41vk_research_12summarizegeminiBLOCK
TRUST ARCHITECTURE
Built for security review, not security theater.
Map virtual-key access, encrypted provider credentials, guardrail decisions, and audit evidence to the controls your enterprise already operates.
SOC 2 control alignment
Architecture supports evidence collection for access, change, and monitoring controls.
ISO 27001 readiness
Map credential, access, policy, and audit practices to ISMS control objectives.
NIST AI RMF mapping
Operational visibility across governed model access, guardrails, usage, and provider outcomes.
Control-alignment statements describe product architecture and are not claims of EnvisionAI certification.
WORKS WITHOpenAIAnthropicGeminiVertex AIGroqOllama
TECHNICAL EVALUATION
Review the control path with your security and platform teams.
Map the architecture to your providers, application clients, governance scopes, data boundaries, and operating requirements.