LLM Providers

Aurora requires an LLM provider for its AI-powered investigation and Root Cause Analysis (RCA) capabilities. You can use a single gateway like OpenRouter, connect directly to individual providers, or run models locally with Ollama.

Supported Providers

Provider	Mode	Environment Variable	Get API Key
OpenRouter	Gateway	`OPENROUTER_API_KEY`	openrouter.ai/keys
OpenAI	Direct	`OPENAI_API_KEY`	platform.openai.com
Anthropic	Direct	`ANTHROPIC_API_KEY`	console.anthropic.com
Google AI	Direct	`GOOGLE_AI_API_KEY`	ai.google.dev
Vertex AI	Direct	`VERTEX_AI_PROJECT` + credentials	console.cloud.google.com
Ollama	Direct	`OLLAMA_BASE_URL`	ollama.com (free, local)

Only one provider is required.

Provider Modes

Aurora supports two routing modes, controlled by LLM_PROVIDER_MODE:

OpenRouter Mode (default)

Routes all LLM requests through OpenRouter, giving you access to multiple model providers with a single API key.

LLM_PROVIDER_MODE=openrouter
OPENROUTER_API_KEY=sk-or-v1-...

Direct Mode

Connects directly to each provider's native API. Use this when running models locally with Ollama or when you prefer direct API access.

LLM_PROVIDER_MODE=direct

In direct mode, Aurora auto-detects the provider from the model name prefix (e.g., anthropic/claude-3-haiku routes to Anthropic, google/gemini-2.5-flash routes to Google AI).

note

Vertex AI and Ollama always use their native SDKs regardless of LLM_PROVIDER_MODE. OpenAI, Anthropic, and Google AI models can all be routed through OpenRouter.

Supported Models

Provider	Model	Notes
OpenAI	`openai/gpt-5.4`	Latest flagship, 1M context
	`openai/gpt-5.2`	Previous flagship
	`openai/o3`	Strong reasoning model
	`openai/o4-mini`	Fast reasoning, lower cost
	`openai/o3-mini`	Compact reasoning model
	`openai/gpt-4.1`	Reliable all-rounder
	`openai/gpt-4.1-mini`	Fast and affordable
	`openai/gpt-4o`	Multimodal (text + vision)
	`openai/gpt-4o-mini`	Cheapest OpenAI option
Anthropic	`anthropic/claude-opus-4.6`	Most capable, 1M context
	`anthropic/claude-sonnet-4.6`	Near Opus quality at lower cost
	`anthropic/claude-opus-4.5`	Previous generation flagship
	`anthropic/claude-sonnet-4.5`	Balanced quality and speed
	`anthropic/claude-haiku-4.5`	Fast, affordable
	`anthropic/claude-3.5-sonnet`	Widely used, reliable
	`anthropic/claude-3-haiku`	Cheapest (default RCA model)
Google Gemini	`google/gemini-3.1-pro-preview`	Latest flagship with thinking
	`google/gemini-3-flash-preview`	Fast, outperforms 2.5 Pro
	`google/gemini-2.5-pro`	Strong for complex tasks
	`google/gemini-2.5-flash`	Cost-effective
	`google/gemini-2.5-flash-lite`	Cheapest Gemini option
Vertex AI	`vertex/gemini-3.1-pro-preview`	Latest flagship with thinking
	`vertex/gemini-3-flash-preview`	Fast, enterprise-grade
	`vertex/gemini-2.5-pro`	Strong for complex tasks
	`vertex/gemini-2.5-flash`	Cost-effective with IAM auth
	`vertex/gemini-2.5-flash-lite`	Cheapest Vertex option
Ollama	`ollama/llama3.1`	Meta's Llama 3.1 (8B/70B)
	`ollama/qwen2.5`	Alibaba's Qwen 2.5 (various sizes)
	Any model via `ollama pull`

Model names use the provider/model format. New models from each provider are generally supported automatically — update your RCA_MODEL or select them in the UI.

Provider Setup

OpenRouter (Recommended for Getting Started)

The easiest way to get started. One API key gives you access to models from OpenAI, Anthropic, Google, Meta, and more.

OPENROUTER_API_KEY=sk-or-v1-...
LLM_PROVIDER_MODE=openrouter

OpenAI

OPENAI_API_KEY=sk-...
LLM_PROVIDER_MODE=direct

Anthropic

ANTHROPIC_API_KEY=sk-ant-...
LLM_PROVIDER_MODE=direct

Google AI (Gemini via API Key)

For using Gemini models with a Google AI Studio API key.

GOOGLE_AI_API_KEY=AIza...
LLM_PROVIDER_MODE=direct

Vertex AI (Gemini via Google Cloud)

For organizations using Google Cloud. Vertex AI provides enterprise-grade access to Gemini models with IAM-based authentication.

Requirements:

A Google Cloud project with Vertex AI API enabled
A service account with the Vertex AI User role

Setup:

# Required: Google Cloud project ID
VERTEX_AI_PROJECT=my-gcp-project

# Required: Service account credentials (JSON string)
VERTEX_AI_SERVICE_ACCOUNT_JSON={"type":"service_account","project_id":"...","private_key":"..."}

# Optional: Location (default: global)
VERTEX_AI_LOCATION=global

LLM_PROVIDER_MODE=direct

Authentication options:

Service account JSON (recommended): Set VERTEX_AI_SERVICE_ACCOUNT_JSON to the full JSON contents of your service account key file.
Application Default Credentials (ADC): If running on GCP (Cloud Run, GKE), ADC is automatic — just set VERTEX_AI_PROJECT.
Credentials file path: Set GOOGLE_APPLICATION_CREDENTIALS to the path of your service account key file.

tip

The project ID is automatically extracted from the service account JSON if VERTEX_AI_PROJECT is not set.

Ollama (Local Models)

Run models locally on your own hardware with Ollama. No API key needed.

Setup:

Install Ollama on your host machine

Pull the models you want to use:

ollama pull llama3.1
ollama pull qwen2.5:32b

Configure Aurora:

OLLAMA_BASE_URL=http://host.docker.internal:11434
LLM_PROVIDER_MODE=direct

Docker networking

host.docker.internal allows Docker containers to reach services running on the host machine. This works out of the box on macOS and Windows. On Linux, Aurora's Docker Compose files include the extra_hosts configuration needed for this to work.

Recommended models for RCA:

Model	Size	Notes
`llama3.1:70b`	70B	Best quality for complex RCA
`qwen2.5:32b`	32B	Good balance of quality and speed
`llama3.2`	3B	Fast, but limited tool calling

RCA Model Configuration

By default, Aurora uses anthropic/claude-3-haiku for background Root Cause Analysis. You can change this to any supported provider/model.

# Format: provider/model-name
RCA_MODEL=anthropic/claude-3-haiku

Examples:

# Anthropic (default)
RCA_MODEL=anthropic/claude-3-haiku

# OpenAI
RCA_MODEL=openai/gpt-4o

# Google AI
RCA_MODEL=google/gemini-2.5-flash

# Vertex AI
RCA_MODEL=vertex/gemini-2.5-flash

# Ollama (local)
RCA_MODEL=ollama/llama3.1

When RCA_MODEL is not set, the default depends on RCA_OPTIMIZE_COSTS:

RCA_OPTIMIZE_COSTS=true (default): Uses anthropic/claude-3-haiku
RCA_OPTIMIZE_COSTS=false: Uses anthropic/claude-opus-4.5

Cost Considerations

LLM costs depend on:

Tokens processed: Longer investigations use more tokens
Model choice: Larger models cost more per token
Frequency: More investigations = higher costs

Cost Optimization

Set RCA_OPTIMIZE_COSTS=true to use cheaper models for background RCA
Use OpenRouter for flexible, pay-per-token pricing
Use Ollama for zero API costs (requires local GPU)

Troubleshooting

"Invalid API key"

Check key is correctly copied (no extra spaces)
Verify key is active in provider dashboard
Ensure correct environment variable name

"Rate limit exceeded"

Wait and retry
Consider upgrading your API tier
Reduce concurrent investigations

"Model not available"

Check provider status page
Try a different model
Ensure your API key has access to the model

Vertex AI: "DefaultCredentialsError"

Verify VERTEX_AI_PROJECT is set
Check that VERTEX_AI_SERVICE_ACCOUNT_JSON contains valid JSON
Ensure the service account has the Vertex AI User role

Ollama: "Provider not available"

Verify Ollama is running: curl http://localhost:11434/api/tags
Check that the model is pulled: ollama list
Ensure OLLAMA_BASE_URL is correct (use host.docker.internal in Docker)

Supported Providers​

Provider Modes​

OpenRouter Mode (default)​

Direct Mode​

Supported Models​

Provider Setup​

OpenRouter (Recommended for Getting Started)​

OpenAI​

Anthropic​

Google AI (Gemini via API Key)​

Vertex AI (Gemini via Google Cloud)​

Ollama (Local Models)​

RCA Model Configuration​

Cost Considerations​

Cost Optimization​

Troubleshooting​

"Invalid API key"​

"Rate limit exceeded"​

"Model not available"​

Vertex AI: "DefaultCredentialsError"​

Ollama: "Provider not available"​