AI gateway for Enterprise teams — route requests to any LLM provider.
You dont need to worry about different API formats, rate limits, or provider outages. Aurora Gateway handles it all for you.
Built forraw speed
Engine: Go-Native / LLM Edge Architecture
Aurora Gateway is engineered in native Go and validated with rigorous side-by-side load tests. We prioritize latency elimination and infinite infrastructure scale above all legacy standards.
BEYOND PERFORMANCE BEYOND PERFORMANCE BEYOND PERFORMANCE BEYOND
OSS Gateway Module
Native Go-Runtime / Zero-trust Auth Layer
47.5ms
8K r/s
99.99%
Multi-provider LLM routing with semantic caching, usage analytics, and provider pools — fully self-hosted and free.
Deploy via Docker, Helm, or binary. MIT-licensed. No usage limits or hidden fees.
Enterprise Node
SAML + RBAC / Multi-Region Deployment
Identity, RBAC, audit logs, budgets, tenant management, and cluster controls for production-scale AI deployments.
SSO, compliance exports, priority support, and advanced observability included with the Enterprise license.
One request.
Six steps. Full pipeline.
Every request that hits Aurora runs through auth, pool selection, cache, upstream, audit, and usage recording. Around that core sit routing, security, caching, and observability features you can mix in or leave off.
The six steps run on every request — streaming-safe, no extra hops, no plugin overhead. Steps 1, 2, 3, 4, 6 are in-process; step 5 is the only network call.
In-process
- ›Auth verify
- ›Pool pick
- ›Cache lookup
- ›Audit + usage
Network
- ›Upstream call
- ›Streaming proxy
- ›SSE relay
Always-on
- ›Token counting
- ›Latency tracking
- ›Error capture
Everything Aurora does.
Mixed in around the core pipeline. Toggle on or off via env, YAML, or admin API.
Routing & Failover
05- Provider Pools
Group compatible upstreams for load balancing
- Combos
Ordered model fallback chains as one name
- Manual Fallback
Per-request rules for which upstream runs next
- Auto Fallback
Degrade automatically on configured failures
- Resilience
Exponential backoff, jitter, circuit breaker
Caching
04- Exact CacheOSS
Hash-matched response reuse, Redis or in-memory
- Semantic CacheOSS
Vector-similarity cache with tunable threshold
- Provider Prompt Cache
Forward cache_control to Anthropic / OpenAI / Gemini
- Model Registry Cache
Refreshed model metadata, offline-safe
Security & Governance
06- Master KeyOSS
Top-level gateway auth for admin operations
- Managed KeysOSS
Scoped, expiring, rate-limited, with usage stats
- RBAC + TenantsEnterprise
Roles, users, budgets per workspace
- Guardrails
Input + output policy through the gateway pipeline
- Audit Log
Request, response, and header history to storage
- Token Saver
Caveman-style output compression for cost control
Observability
04- Usage Analytics
Requests, tokens, cost, by model, provider, user, range
- Prometheus
Standard /metrics endpoint for any scraper
- Admin Dashboard
Providers, pools, keys, guardrails, audit, cache
- Console Streams
Real-time request stream in the dashboard
Storage & Deployment
04- SQLite / Postgres / Mongo
Same API, swap backends via STORAGE_TYPE
- YAML + .env
Config files merged with env vars at startup
- Edition Profiles
Pre-built configs in configs/editions/
- Helm Chart
Production K8s deploy with values-driven config
Developer Surface
03- OpenAI-compatible
Chat, responses, embeddings, models, files, batches
- Anthropic-compatible
/v1/messages, max_tokens auto-injected
- Native Passthrough
/p/{provider}/... for full upstream API access
One curl.
Two formats.
Aurora preserves the exact wire format from each SDK — OpenAI clients see OpenAI responses, Anthropic clients see Anthropic responses.
Aurora$ curl -X POST http://localhost:8080/v1/chat/completions \
-H "Authorization: Bearer $AURORA_MASTER_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "openai/gpt-5.6-terra",
"messages": [{"role": "user", "content": "Hi there!"}]
}'14 Provider Types
Zero Overhead
Native Integration Layer: Anthropic / Mistral / Llama-3 / Cohere / Perplexity AI / Vertex