Route, govern, and observe every LLM call through one OpenAI-compatible endpoint — at sub-millisecond overhead. Proven gateway engineering, now for AI and agents.
Start free, then grow on the managed cloud as your traffic scales — or self-host the open-source core anytime.
Free forever — fully managed, zero ops.
Past 100K, traffic keeps flowing — extra requests just aren't recorded.
Start freeManaged governance for production teams.
$100 per additional 1M, never blocked. Sustained over 5M/mo moves to Enterprise.
Get startedProduction AI at scale, with enterprise compliance.
Custom quota, overage, deployment and SLA.
Talk to salesAny call routed and logged through the gateway, including errors. One streaming response counts as a single request.
On Developer, traffic keeps flowing past 100K — those extra requests simply aren't recorded or counted. No surprise cut-offs.
Past 1M, we auto-add $100 per additional 1M on your monthly invoice. Traffic is never interrupted.
Swipe to compare plans →
| Feature | Developer$0 | Team$149/mo | EnterpriseCustom |
|---|---|---|---|
| Usage & limits | |||
| Recorded requests / month | 100K | 1M | 10M+ |
| Members | 1 | 25 | Unlimited |
| Environments | 1 | 3 | Unlimited |
| Data planes | 1 | 3 | Unlimited |
| Log retention | 3 days | 30 days | Custom |
| Metrics retention | 30 days | 90 days | Custom |
| Monthly overage | Not recorded | $100 / 1M | Custom |
| AI gateway core | |||
| Universal OpenAI-compatible API | |||
| Chat completions + streaming (SSE) | |||
| Anthropic-compatible /v1/messages | |||
| Embeddings · rerank | |||
| Audio (speech-to-text, TTS) | |||
| Image generation | |||
| Provider passthrough | |||
| Playground | |||
| Single-model direct | |||
| Virtual / routing models | — | ||
| Model providers | |||
| OpenAI · Anthropic · Gemini · DeepSeek | |||
| 20+ popular providers | |||
| 100+ via OpenAI-compatibility | |||
| AWS Bedrock · Azure OpenAI · GCP Vertex AI | — | — | |
| Routing & reliability | |||
| Weighted load balancing | — | ||
| Automatic retry | — | ||
| Fallback on errors / 429 | — | ||
| Upstream health checks | — | ||
| Semantic routing | — | — | |
| Cost-optimal routing | — | — | |
| Rate limiting | |||
| Request rate limits (RPM / RPD) | — | ||
| Token rate limits (TPM / TPD) | — | ||
| Concurrency limits | — | ||
| Per-team / per-member limits | — | ||
| Cost & budgets | |||
| Per-key budgets (hard-stop / warn) | — | ||
| Org / env / provider budgets | — | ||
| Budget alerts (75 / 90 / 100%) | — | ||
| Per-model cost tracking | — | ||
| Caching | |||
| Response cache (exact-match) | — | ||
| Memory + Redis backends | — | ||
| Cost-saved telemetry | — | ||
| Semantic cache | — | — | |
| Security & guardrails | |||
| Keyword / regex guardrails | — | ||
| Cloud safety guardrails | — | — | |
| PII redaction · content moderation | — | — | |
| Custom guardrail hooks | — | — | |
| Observability | |||
| Request access logs | |||
| Usage & cost analytics dashboard | Basic | ||
| Prometheus metrics | — | ||
| OpenTelemetry trace export | — | ||
| Alerts | — | ||
| Data Lake / bucket export | — | — | |
| Access control & org | |||
| API key management + rotation | |||
| Virtual keys & model allowlist | |||
| Personal access tokens (CLI / CI) | |||
| Social login (GitHub / Google) | |||
| RBAC | — | ||
| Teams | — | ||
| Organization management | — | — | |
| SSO (SAML / OIDC) | — | — | |
| Audit logs | — | — | |
| Deployment & compliance | |||
| Managed SaaS hosting | |||
| Provider-key encryption at rest | |||
| Self-host / on-prem option | — | — | |
| Private / VPC deployment | — | — | |
| SOC 2 Type II · ISO 27001 · GDPR · HIPAA | — | — | |
| BAA · data isolation | — | — | |
| Support & SLA | |||
| Support | Community | Standard | Dedicated |
| SLA | — | Standard | Custom |
Run AISIX in your own cloud or VPC, enforce org-wide policy, and meet the compliance bar — backed by a dedicated team and an SLA.
No. You bring your own provider keys and pay your LLM bill directly to OpenAI, Anthropic, and others. AISIX Cloud only charges the plan subscription — there is no markup on your tokens.
Recorded requests — any call routed and logged through the gateway, including error responses. One streaming response counts as a single request.
On Developer, your traffic keeps flowing past 100K — those extra requests simply aren't recorded, so they don't show up in your logs or analytics, and nothing is ever blocked. On Team, we automatically add $100 per additional 1M requests — traffic is never interrupted. Sustained usage over 5M/mo is a good point to move to Enterprise.
Yes. AISIX is open source under Apache-2.0 — run the full gateway as a single Rust binary, free, with community support. The managed cloud adds the hosted control plane, dashboard, budgets, RBAC, and SLAs on top. Enterprise can also run the managed stack inside your own cloud / VPC.
More than 100 providers through one OpenAI-compatible API, including 20+ popular integrations (OpenAI, Anthropic, Gemini, DeepSeek, Groq, Mistral, Cohere, Qwen, Together, Fireworks, and more). Cloud-hosted providers — AWS Bedrock, Azure OpenAI, and GCP Vertex AI — are available on Enterprise.
Organization management, SSO (SAML / OIDC), audit logs, and SOC 2 Type II / ISO 27001 / GDPR / HIPAA are part of Enterprise. Talk to sales to scope your requirements and deployment model.
On AISIX Cloud, provider keys are envelope-encrypted at rest and decrypted only at request time; each data plane is scoped to its own environment keyspace. On self-host, all data and keys stay entirely within your own infrastructure.
The data plane is a native Rust proxy with sub-millisecond overhead, so the gateway is negligible next to LLM inference time.
OpenAI-compatible — your existing code works unchanged. Start free, no credit card.