Choose a client API and an upstream first, then add routing and policy. These pages describe Development (current main). Build features and compatibility can differ from released v0.10.0; use the documentation snapshot matching your binary. See the support audit for the reviewed revision and validation evidence.
- Quickstart: build, validate, and forward your first hosted request.
- Choose a client and provider: protocol, credentials, feature selection, and compatibility limits in one table.
- Operator recipes: prerequisites, full configurations, startup, requests, expected results, and cleanup.
- Chat Completions: native OpenAI-compatible forwarding.
- Anthropic Messages: native forwarding, translation, SDK versions, and backend qualification.
- Responses: distinguish native passthrough from experimental gateway-managed state and tool execution.
- Conversations: local experimental CRUD, storage, and continuation boundaries.
- Translation compatibility: supported, lossy, rejected, and known-gap behavior by direction.
- Provider selection: OpenAI-compatible, Anthropic, Azure, Vertex, and Bedrock paths.
- Credential boundaries: caller keys, gateway authentication, backend secrets, and callout credentials.
- Hosted OpenAI, native or translated Anthropic, and Azure, Vertex, and Bedrock adapters.
- Model aliases and provider routing: start with a static mapping, then choose dynamic selection, failover, or session affinity.
- Routing facts: which component owns each fact.
- Routing examples: complete YAML chains.
- Guardrails and token usage: external screening, usage extraction, streaming limits, and metrics.
- Operator capability overview: identity, token limits, metering, credentials, and required filter order.
- Outbound callout security.
- Experimental managed Responses and tools.
- MCP and A2A: protocol routing and brokering are distinct from model-driven execution.
- Files and stores, retained-payload budget, and store migrations.
- Operations checklist and troubleshooting: deployment, configuration validation, observability, and upgrades.
- Capacity planning: file descriptors and admission.
- FIPS deployment and OpenTelemetry.
- Security policy.
- Feature and build reference.
- Generated filter reference and example catalog.
- Development setup, adding filters, and embedding filters.
- CLI clients with vLLM, OpenShell, qualification, and GPU testing.