Start on one host. Scale to governed enterprise AI.

Yes. Evaluation Edition provides a 30-day self-hosted evaluation at no charge. It can be deployed with Docker Compose so your team can validate gateway integration, virtual keys, provider routing, budgets, rate limits, guardrails, and audit visibility using your own applications and provider accounts. Assisted remote evaluation is also available for $5,000 and is credited toward an annual license when purchased within 60 days.

Every edition runs in infrastructure controlled by your organization. Evaluation Edition supports a single-host Docker Compose deployment, while Standard Edition licenses a customer-managed production deployment. Professional Edition supports multi-replica and Kubernetes deployments using Helm, with shared runtime state so rate limits and containment hold across replicas.

Agent Access Manager is licensed through an annual software subscription rather than a month-to-month hosted-service plan. Standard Edition is $30,000 per year per production deployment and includes one non-production deployment, updates, security patches, and Standard Support. Professional Edition starts at $90,000 per year and is scoped according to production clusters, environments, identity integrations, operational requirements, and support needs. Customer infrastructure and AI-provider usage charges are separate from the software subscription.

Agent Access Manager supports OpenAI-compatible endpoints and provider adapters for Anthropic, Gemini, and Vertex AI, along with supported self-hosted model endpoints. Applications and frameworks that can use a configurable OpenAI-compatible base URL can connect through scoped virtual keys and model aliases. Provider usage remains in the customer’s existing provider accounts.

Standard Edition licenses a single-instance production deployment and includes the complete governance feature set: virtual keys, routing, budgets, rate limits, guardrails with AI classifiers, searchable audit, detections, behavioral analytics, and SOAR containment. Professional Edition adds the scale-out architecture: multiple application replicas behind a load balancer, Kubernetes deployment with Helm and horizontal autoscaling, shared runtime state so rate limits and containment hold cluster-wide, integration with external managed PostgreSQL and OpenSearch, S3-compatible report storage, custom identity-provider architecture, and Priority Support.

The MCP Gateway governs agent tool calls to MCP servers with the same guardrail, policy, and audit pipeline as model traffic. The MCP Registry provides a private catalog of approved remote and self-hosted MCP servers with per-agent, per-key, and per-team tool permissions. MCP Orchestration, included in Professional Edition, additionally deploys self-hosted MCP servers as isolated workloads and manages their configuration, secrets, logs, status, and restarts. The MCP Gateway and MCP Registry are available in every edition.

Provider credentials are encrypted with AES-256-GCM and accessed only by the gateway when executing authorized requests. Agents and applications use scoped virtual keys — stored only as hashes — instead of receiving reusable provider credentials. Agent Access Manager, its policy configuration, and its audit records run in the customer’s environment, while model inputs are sent only to the providers or endpoints the customer configures.

Agent Access Manager can enforce provider, model, budget, RPM, TPM, content, PII, and secret policies at runtime. Configured request and response guardrails can allow, flag, redact, or block activity, including tool-call arguments and streamed responses. Budgets and rate limits are checked before any provider is called and are enforced across the organization, team, project, and key at once. Every request is recorded with the applicable identity, policy decision, and outcome for investigation and operational review.

Standard Edition remote onboarding is available for a one-time fee of $7,500. Professional Edition remote onboarding starts at $20,000 and is scoped according to deployment complexity, Kubernetes architecture, identity integration, and operational readiness. A typical engagement can include architecture review, deployment guidance, provider integration planning, policy configuration, validation, and administrator handoff. The final scope, schedule, responsibilities, and deliverables are confirmed before the engagement begins.

Yes. Onsite professional services are available at $2,500 per consultant-day with a three-day minimum. Travel, lodging, visa costs, and local transportation are billed separately. International travel days are billed at 50% of the consultant-day rate, and the engagement scope is agreed in advance.

Standard Edition includes Standard Support with 8x5 local business-hours coverage and initial response targets of 4 business hours for P1, 1 business day for P2, and 2 business days for P3. Professional Edition includes Priority Support with 24x7 P1 coverage and initial response targets of 2 hours for P1, 4 business hours for P2, and 1 business day for P3. Business Critical Support is an optional Professional Edition add-on starting at $30,000 per year or 25% of annual license value, whichever is greater. It provides 24x7 P1 and P2 coverage with initial response targets of 1 hour for P1, 2 hours for P2, and 4 business hours for P3. These are initial response targets, not guaranteed resolution times or customer-infrastructure uptime commitments.