Skip to main content

LLM gateway policy and audit

Apply governance policies to the model calls that flow through the LLM gateway, and review what gets logged. Policies let you control which models agents can use, flag unusual usage patterns, and maintain an audit trail for compliance.

Model governance

The LLM gateway routes model calls to Anthropic's API on behalf of your organization. You can restrict which models agents are allowed to use from the Policy tab in LLM Gateway settings.

SettingEffect
Allowed modelsAgents can only request models on the allowlist. Requests for other models are rejected with a clear error.
Default modelRequests that don't specify a model are routed to this model. Useful for ensuring cost predictability across agents.

Coworker Agent harness configuration sets the model at the agent level; the gateway-level allowlist is an outer bound that prevents any agent, including a misconfigured or compromised one, from calling a model outside your policy.

Rules

LLM gateway traffic supports the same rule system used for MCP tool call governance. Rules run on each model request and can flag, block, or route the call based on content or metadata.

Common uses for LLM gateway rules:

  • Flag high-token requests: alert when a single request exceeds a token threshold (useful for catching runaway loops).
  • Block sensitive content: stop requests that contain credentials or PII before they reach the model.
  • Notify on anomalous usage: send a Slack alert when usage spikes above a baseline.

Create and manage rules at app.mintmcp.com/monitor/rules. See Agent Monitor rules for the full rule authoring reference.

Audit logging

Every model call through the LLM gateway is logged with:

FieldDescription
Agent / keyWhich gateway key made the call, attributed to the agent or developer it was issued for
ModelThe model requested and the model actually used
TokensInput and output token counts
TimestampWhen the request was made and when it completed
StatusSuccess, blocked by policy, or error

Logs are available in LLM GatewayActivity and can be exported for SIEM integration. See SIEM export.

Usage limits

Per-key and per-organization usage limits prevent runaway cost from a misconfigured agent. Configure limits in LLM GatewayPolicy:

Limit typeScopeEffect when exceeded
Daily token budgetPer keyRequests from that key are blocked until the budget resets
Request rate limitPer keyRequests are throttled with a 429 Too Many Requests response
Org monthly budgetOrganizationAll gateway calls are blocked until an admin raises the budget or the month resets

Permissions

Managing LLM gateway policies requires the LLM gateway manage permission. Members without it can view their own usage but cannot change policy settings or issue keys.

Assign the permission through Roles and permissions.

Next steps