AISIX
네이티브 AI Gateway
LLM 및 AI 에이전트용

AISIX is fully open-source (Apache 2.0) and built with Rust for sub-millisecond proxy overhead. Unlike proprietary alternatives, AISIX unifies traditional API traffic and AI/LLM traffic in a single gateway, supports 100+ LLM providers through an OpenAI-compatible API, and provides enterprise features like token rate limiting, prompt guardrails, PII redaction, and comprehensive audit logging.

Yes. AISIX supports 100+ LLM providers including OpenAI, Anthropic Claude, Google Gemini, DeepSeek, Mistral, Qwen, Amazon Bedrock, Azure OpenAI, and more. All providers are accessed through a single OpenAI-compatible API, so you can switch or load-balance across providers without changing your application code.

Yes. The AISIX open-source edition is free forever under the Apache 2.0 license. It includes multi-LLM load balancing, token rate limiting, virtual key management, OpenTelemetry observability, and model guardrails. An Enterprise edition with RBAC, audit logs, priority support, and custom SLAs is available for teams managing AI traffic at scale.

You can deploy AISIX in minutes using Docker, Kubernetes (Helm chart), or build from source. Visit the AISIX documentation at docs.api7.ai/aisix for quickstart guides, API references, and deployment tutorials. The OpenAI-compatible API means your existing LLM integrations work with zero code changes.

AISIX supports both token-based (TPM) and request-based (RPM) rate limiting. You can set limits per consumer, per route, per model, or cluster-wide. It also supports concurrency limits to prevent any single consumer from monopolizing capacity. All rate limiting policies can be updated dynamically via the Admin API without restarts.

AISIX provides multiple security layers: prompt injection detection blocks malicious inputs, content moderation filters inappropriate content, PII redaction removes sensitive data before it reaches the model, and comprehensive audit logging tracks every request for regulatory compliance. Per-key access controls restrict which models each API key can call.