Portkey

97%

COVERAGE

Multi-Provider Support

Breadth of LLM providers supported through a single unified API — OpenAI, Anthropic, Google, AWS, Azure, Cohere, Mistral, open-source, and more.

Smart Routing

Intelligent request routing based on cost, latency, model capability, or custom logic. Includes latency-based, cost-optimized, conditional, and semantic routing.

Fallback & Retry

Automatic failover to backup providers/models on failure. Retry logic with exponential backoff, circuit breaking, and health-aware routing.

Load Balancing

Distribute requests across endpoints. Round-robin, weighted, least-connections, and performance-aware strategies.

Semantic Caching

Cache LLM responses and serve semantically similar requests from cache. Reduces cost and latency for repeated queries.

Cost Tracking & Budgets

Real-time token cost tracking per user, team, project, or API key. Budget limits, spend alerts, and cost attribution.

Rate Limiting & Quotas

Token-aware and request-based rate limiting. Per-user, per-team, per-key quotas.

Latency Performance

Gateway overhead added to requests. Sub-ms is ideal. Rust/Go implementations outperform Python-based gateways.

Guardrails & Safety

Built-in content filtering, PII detection, toxicity blocking, prompt injection defense, and output validation.

Auth & RBAC

Authentication (API keys, OAuth, SSO), role-based access control, workspace isolation, fine-grained permissions.

Audit Logging

Immutable logs of all requests, responses, routing decisions, and policy violations. Compliance-exportable.

MCP Server Support

Model Context Protocol server support — enabling AI agents to access external tools through the gateway with auth and governance.

Observability

Dashboards for request volume, latency, errors, tokens, model performance. Exportable to Datadog, Grafana, etc.

Prompt Management

Version control, deploy, A/B test prompts as gateway assets. Templates, variable injection, environment promotion.

Self-Hosted / OSS

Deploy on own infrastructure. Open-source, Docker/K8s, air-gapped, data residency compliance.

Streaming & SSE

Full streaming response support (SSE), chunked transfer, real-time token delivery for chat UIs.