OSS / Getting Started

Quick Start

Get Aurora running in five minutes: install, set one provider key, send your first request, open the dashboard, and issue a managed key.

Browse docs

Aurora is a single OpenAI- and Anthropic-compatible gateway. Install it, set one or more provider keys, and your existing SDK calls start working against http://localhost:8080/v1 instead of going directly to a provider. From there you can add caching, guardrails, pools, and managed keys without touching application code.

This page walks you from zero to a working request in under five minutes.

Install & start Aurora

Pick whichever method matches how you deploy. All options expose the same /v1 endpoints and /admin/dashboard.

One command — installs, scaffolds config, and starts:

text

npx -y iaurora

What happens:

Downloads the native aurora binary
Creates config.yaml, .env, and data/ in the current directory
Starts the gateway on port 8080

Stop it with Ctrl-C. If you restart, the same .env and config.yaml are reused.

Before your first request — stop the gateway (Ctrl-C), set at least one provider key in .env, then run aurora again:

text

# ── Required: set at least one ────────────────────
OPENAI_API_KEY="sk-..."
ANTHROPIC_API_KEY="sk-ant-..."
# ── (gemini, groq, deepseek, xai, etc. — same pattern)
 
# ── Optional: secure your gateway ─────────────────
AURORA_MASTER_KEY="your-secure-key"

Then restart with:

text

aurora

Confirm Aurora is running

In another terminal, list the models Aurora discovered:

text

curl http://localhost:8080/v1/models \
  -H "Authorization: Bearer your-master-key"

If you set OPENAI_API_KEY, you should see gpt-5.6-sol, gpt-5.6-luna, etc. If the response is data: [], no provider keys are set — check your .env and restart.

Send your first request

Pick a model from the list /v1/models returned (e.g. openai/gpt-5.6-luna) and send a chat completion:

The Anthropic SDK path requires ENABLE_ANTHROPIC_INGRESS=true — see Anthropic Messages API.

Open the admin dashboard

Open http://localhost:8080/admin/dashboard in a browser. The dashboard is a React SPA embedded in the binary. It shows providers, models, aliases, pools, guardrails, usage, audit logs, and a live console stream.

Admin API endpoints live under /admin/api/v1/* — see Admin API.

Issue a managed key for your team

The master key is a bootstrap secret. For day-to-day use, create scoped managed keys:

text

curl -X POST http://localhost:8080/admin/api/v1/auth-keys \
  -H "Authorization: Bearer your-master-key" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "team-alpha",
    "provider_models": [{"provider": "openai", "models": ["gpt-5.6-luna"]}],
    "rate_limit": {"requests_per_minute": 60}
  }'

The response contains a key value — hand that to your teammate. They can call Aurora the same way you did above. See Managed API Keys.

Optional: features to turn on next

These are off by default. Enable them when you need them.

Feature	Toggle	Why
Audit logging	`LOGGING_ENABLED=true`	Inspect every request/response in the dashboard
Prometheus metrics	`METRICS_ENABLED=true`	Scrape `/metrics` for dashboards and alerting
Guardrails	`GUARDRAILS_ENABLED=true`	PII redaction, regex blocking, prompt injection protection
Response cache (exact)	`RESPONSE_CACHE_SIMPLE_ENABLED=true` + `REDIS_URL`	SHA-256 keyed cache, `X-Cache: HIT` on hits
Response cache (semantic)	`SEMANTIC_CACHE_ENABLED=true` + vector store	Hit on similar prompts via Qdrant/pgvector/Pinecone/Weaviate
Token saver	`TOKEN_SAVER_ENABLED=true`	Compress tool output and request concise completions
Anthropic Messages ingress	`ENABLE_ANTHROPIC_INGRESS=true`	Expose `/v1/messages` for native Anthropic SDK

Next steps

These five links cover the next hour of typical onboarding. Skim them in order; each builds on the previous.

Concepts — five-minute glossary: alias vs combo vs pool vs override vs workflow.
Providers Overview — pick the providers you want and skim their one-page guides.
Managed API Keys — give your team scoped, rate-limited keys.
Cache — turn on the response cache and start saving tokens.
Guardrails — add PII redaction and regex blocking before you go to production.

Reference

Editions — what is OSS-only and what needs Enterprise
Configuration — full env-var and YAML reference
Deployment Overview — Docker Compose, Kubernetes, source builds

← All docs

OSS / Getting Started

Quick Start

Get Aurora running in five minutes: install, set one provider key, send your first request, open the dashboard, and issue a managed key.

Browse docs

This page walks you from zero to a working request in under five minutes.

Install & start Aurora

Pick whichever method matches how you deploy. All options expose the same /v1 endpoints and /admin/dashboard.

One command — installs, scaffolds config, and starts:

text

npx -y iaurora

What happens:

Downloads the native aurora binary
Creates config.yaml, .env, and data/ in the current directory
Starts the gateway on port 8080

Stop it with Ctrl-C. If you restart, the same .env and config.yaml are reused.

Before your first request — stop the gateway (Ctrl-C), set at least one provider key in .env, then run aurora again:

text

# ── Required: set at least one ────────────────────
OPENAI_API_KEY="sk-..."
ANTHROPIC_API_KEY="sk-ant-..."
# ── (gemini, groq, deepseek, xai, etc. — same pattern)
 
# ── Optional: secure your gateway ─────────────────
AURORA_MASTER_KEY="your-secure-key"

Then restart with:

text

aurora

Confirm Aurora is running

In another terminal, list the models Aurora discovered:

text

curl http://localhost:8080/v1/models \
  -H "Authorization: Bearer your-master-key"

If you set OPENAI_API_KEY, you should see gpt-5.6-sol, gpt-5.6-luna, etc. If the response is data: [], no provider keys are set — check your .env and restart.

Send your first request

Pick a model from the list /v1/models returned (e.g. openai/gpt-5.6-luna) and send a chat completion:

The Anthropic SDK path requires ENABLE_ANTHROPIC_INGRESS=true — see Anthropic Messages API.

Open the admin dashboard

Admin API endpoints live under /admin/api/v1/* — see Admin API.

Issue a managed key for your team

The master key is a bootstrap secret. For day-to-day use, create scoped managed keys:

text

curl -X POST http://localhost:8080/admin/api/v1/auth-keys \
  -H "Authorization: Bearer your-master-key" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "team-alpha",
    "provider_models": [{"provider": "openai", "models": ["gpt-5.6-luna"]}],
    "rate_limit": {"requests_per_minute": 60}
  }'

The response contains a key value — hand that to your teammate. They can call Aurora the same way you did above. See Managed API Keys.

Optional: features to turn on next

These are off by default. Enable them when you need them.

Feature	Toggle	Why
Audit logging	`LOGGING_ENABLED=true`	Inspect every request/response in the dashboard
Prometheus metrics	`METRICS_ENABLED=true`	Scrape `/metrics` for dashboards and alerting
Guardrails	`GUARDRAILS_ENABLED=true`	PII redaction, regex blocking, prompt injection protection
Response cache (exact)	`RESPONSE_CACHE_SIMPLE_ENABLED=true` + `REDIS_URL`	SHA-256 keyed cache, `X-Cache: HIT` on hits
Response cache (semantic)	`SEMANTIC_CACHE_ENABLED=true` + vector store	Hit on similar prompts via Qdrant/pgvector/Pinecone/Weaviate
Token saver	`TOKEN_SAVER_ENABLED=true`	Compress tool output and request concise completions
Anthropic Messages ingress	`ENABLE_ANTHROPIC_INGRESS=true`	Expose `/v1/messages` for native Anthropic SDK

Next steps

These five links cover the next hour of typical onboarding. Skim them in order; each builds on the previous.

Concepts — five-minute glossary: alias vs combo vs pool vs override vs workflow.
Providers Overview — pick the providers you want and skim their one-page guides.
Managed API Keys — give your team scoped, rate-limited keys.
Cache — turn on the response cache and start saving tokens.
Guardrails — add PII redaction and regex blocking before you go to production.

Reference

Editions — what is OSS-only and what needs Enterprise
Configuration — full env-var and YAML reference
Deployment Overview — Docker Compose, Kubernetes, source builds