Datadog AI Gateway lets you govern how your organization uses large language models (LLMs) by routing agent and application traffic through a single endpoint. You can set budgets that stop spend when it hits a limit, decide which teams can reach which models, and enforce security and privacy policies on requests as they pass through. The gateway routes each call to the right model and falls back when one is over budget, rate-limited, or down. It also shifts traffic to cheaper models without any team rewriting code.
AI Gateway works with OpenAI, Anthropic, Amazon Bedrock, Google Vertex AI, and Azure OpenAI Service. Whether your engineers are running coding agents like Claude Code, Codex, or Pi, or you’re shipping agentic applications to production, the gateway connects to your own provider accounts and requires no change to how your teams call models. This Product Preview also includes the platform connections behind it: traces in Datadog Agent Observability, spend in Datadog Cloud Cost Management, and AI Guard to detect and block prompt injection and other attacks.
To qualify for this Product Preview, you should already be sending production LLM traffic and be ready to route some of it through the gateway.