Available as a standalone binary or for Kubernetes, it routes, secures, observes, and governs HTTP, gRPC, TCP, LLM, MCP, and agent-to-agent traffic without requiring separate gateways.

Its LLM Gateway connects applications to providers including OpenAI, Anthropic, Amazon Bedrock, Gemini, and Vertex, as well as self-hosted models. Features include credential management, failover, token budgets, semantic caching, prompt redaction, and inference routing for vLLM, TGI, and Triton. Native MCP and A2A gateways support tool discovery, access controls, auditing, and communication among agent frameworks and runtimes.

For APIs and services, agentgateway provides TLS and mTLS, OIDC and JWT authentication, authorization policies, API keys, rate limiting, and external authorization. Observability tools include OpenTelemetry, Prometheus metrics, Jaeger traces, request logs, and a command-line interface for inspecting live traffic and configuration. Additional capabilities cover prompt and tool-call guardrails, cost attribution, and spending limits. The project also offers documentation, an admin-based LLM playground, and community channels.