Unified model access
Expose one OpenAI-compatible endpoint for OpenAI, Anthropic, Gemini, Azure OpenAI, AWS Bedrock, Mistral, Groq, and compatible providers.
Enterprise AI gateway
Govern every model, tool, and AI agent
Tygress is a self-hosted enterprise AI gateway in Rust. Route across LLM providers, secure prompts and data, control tokens and spend, govern MCP tools and A2A agents, and apply the same enterprise policy to every API.
Tygress is currently in development. Product capabilities describe the planned launch scope.
ProvidersOpenAIAnthropicGeminiAzure OpenAIAWS BedrockMistralGroqOpenAI-compatible
Every workload
REST/api/v1/orders
gRPCpayments.Charge
LLM/v1/chat/completions
MCPtools/call
One Rust data plane
Tygress
One route model and one policy chain for API, model, tool, and agent traffic.
IdentityRate limitsAI guardrailsCost budgetsAuditOpenTelemetry
Every upstream
APIs & microservices
OpenAI & Anthropic
Bedrock & Gemini
MCP & A2A agents
One enforcement point
AI governance belongs in the request path
Dashboards can report what happened. Tygress enforces what is allowed before the request reaches a provider, tool, or agent: who can call it, what data may leave, how much it may cost, and whether a human must approve it.
Keys and model policy
Issue virtual keys, keep provider credentials server-side, rotate across key pools, and enforce per-consumer model and tool access.
Cost and quota control
Track prompt, completion, cached, and reasoning tokens. Enforce token windows and USD budgets by consumer, route, tenant, provider, or model.
AI security and DLP
Detect prompt injection and jailbreaks, redact PII and secrets in requests or streamed responses, and keep matched data away from providers.
Reliability and caching
Fail over across providers and keys, route by priority or weight, use upstream quota signals, and replay semantically similar answers from cache.
MCP and agent governance
Aggregate MCP servers, enforce tool allowlists, route A2A calls by skill, and hold sensitive actions in a gateway approval queue with audit and webhooks.
Shared operating model
One gateway, four teams, one source of truth
Developers
Use one OpenAI-compatible contract and stable model aliases instead of provider-specific credentials and SDK logic.
Platform teams
Operate model, API, MCP, and A2A traffic through the same route model, deployment topology, and telemetry stack.
Security teams
Enforce identity, data handling, tool access, prompt policy, approvals, and audit before a request leaves the network.
FinOps teams
Attribute tokens and spend to consumers, cap budgets in USD, track cache savings, and observe provider failover and quota state.
More than an LLM proxy
Use the same controls for REST and gRPC
Tygress includes the traffic management, authentication, protocol support, resilience, and extensibility expected from an enterprise API gateway.
FAQ
Enterprise AI gateway questions
What is an enterprise AI gateway?
An enterprise AI gateway is the policy and operations layer between applications and AI systems. It centralizes model access, provider credentials, routing, failover, token and cost controls, prompt security, data-loss prevention, observability, MCP tool access, and agent governance. Tygress combines those controls with a full enterprise API gateway.
How is an AI gateway different from an LLM proxy?
An LLM proxy primarily normalizes model APIs and routes model requests. An enterprise AI gateway adds organizational identity, tenant policy, security controls, audit, cost governance, high availability, and agent protocols. Tygress also carries REST, gRPC, and WebSocket traffic, reducing the need for a separate API gateway.
Which AI providers does Tygress support?
Tygress supports OpenAI, Anthropic, Google Gemini, Azure OpenAI, AWS Bedrock, Mistral, Groq, and any OpenAI-compatible endpoint. Bedrock supports both OpenAI-compatible access and the native Converse API with AWS SigV4.
Can Tygress govern MCP servers and AI agents?
Yes. Tygress can aggregate MCP servers behind one endpoint, generate toolsets from OpenAPI, apply per-consumer tool allowlists, and serve REST upstreams as MCP tools. Its A2A gateway federates agents and can require human approval for sensitive MCP or A2A calls.
Can the AI gateway be self-hosted or air-gapped?
Yes. Tygress is designed to run in a private VPC, on-premises, Kubernetes, or a fully air-gapped environment. Provider credentials, prompts, responses, identity, and policy remain inside the infrastructure you control.
Is Tygress a LiteLLM alternative?
Yes, for teams evaluating a deployed LLM gateway. Tygress provides an OpenAI-compatible model layer with routing, failover, keys, budgets, guardrails, and observability, then adds enterprise API gateway and agent-governance capabilities. Tygress is still pre-launch; LiteLLM is available today.
Control AI before it leaves your network
Build on one enterprise AI gateway
Join the waitlist for early access, deployment guidance, and launch updates.