The control plane for production AI systems.

Yes. We support OpenAI, Anthropic, Azure OpenAI, Cohere, Mistral, and any model accessible via API. You can also bring your own fine-tuned SLMs. Switching models requires no code changes—just update your configuration.

Yes. Our Regulated tier includes full on-premises deployment with air-gapped licensing. We also offer VPC deployment for Enterprise customers who need network isolation without managing infrastructure.

You define eval suites with custom or built-in evaluators (accuracy, latency, cost, safety). Evals run automatically on every deploy. If metrics regress beyond thresholds you set, the deploy is blocked until reviewed. Shadow testing lets you compare new versions against production before switching.

Team plans retain logs for 7 days. Enterprise retains for 90 days. Regulated plans support 1+ year retention with immutable audit logs. You can configure custom retention per data type and export logs to your own storage at any time.

Yes. We support real-time streaming to Splunk, Datadog, and any SIEM that accepts webhooks or syslog. Security events, access logs, and policy violations can all be routed to your existing security infrastructure.

Each data source connector inherits permissions from your identity provider (Okta, Azure AD, etc.). When a user queries RAG, we enforce row-level and document-level access based on their identity. Admins can also define explicit allow/deny rules per connector.