AI Gateway Policies and Profiles

• Budget Limits — Financial guardrails that cap spending per workspace. You can set cost-based limits (in USD) or token-based limits (maximum tokens consumed). When the limit is reached, the associated key is expired to prevent further spending. Budget limits support alert thresholds for advance notification and periodic reset options (weekly or monthly) for recurring workloads. For fine-grained control scoped per user, API key, model, provider, or metadata condition, use policy-based budget rules configured through the Admin API.
• Rate Limits — Velocity controls that cap how many requests a workspace can make per minute and per day. Rate limits prevent runaway usage from exhausting provider quotas and protect other workspaces from being affected by a single high-traffic consumer.
• Configuration — Reusable routing profiles (configs) that define how the gateway routes requests. A config can specify load balancing across multiple providers, automatic fallback to a secondary model when the primary is unavailable, response caching to reduce costs and latency, and inline content policies. Configs can be attached to an API key so that the routing profile applies to every request authenticated with that key.
• Access Policies — Fine-grained authorization rules that control which models, providers, and resources a workspace or API key is permitted to access. Access policies let you enforce model allow- and deny-lists at the policy level, independently of the catalog's model provisioning settings.
• Guardrails — Inline content policies that run on every request and response. Guardrails detect and block prompt injection attempts, identify sensitive data such as PII or credentials, enforce topic restrictions, and can invoke the Prisma AIRS AI Runtime Security inspection policy for deeper threat analysis. Guardrails execute before traffic reaches the provider and before responses reach the caller, so threats are blocked without any post-processing step.
• Agent Endpoint — Rules that govern AI Gateway's Agent Gateway capability. Agent endpoint policies control how autonomous agents authenticate when calling other agents or MCP servers through the gateway, and define what operations each registered agent is permitted to perform.