AI Providers

☁️ Cloud Providers

Cloud providers host models on their infrastructure. You need an API key to use them:

Provider	Best For	Pricing
OpenAI	General purpose, code generation	Pay per token
Anthropic	Complex reasoning, long context	Pay per token
Google Gemini	Multimodal, large context windows	Pay per token
DeepSeek	Code-focused, cost-effective	Pay per token
Minimax	Chinese language tasks	Pay per token
Qwen	Chinese language, reasoning	Pay per token
Aura Cloud	Managed service with E2EE sync	Subscription

🏠 Local Providers

Local providers run models on your own hardware. No API keys, no internet required, complete privacy:

Provider	Setup	Best For
Ollama	ollama pull llama3	Easy local model management
LM Studio	Download from lmstudio.ai	GUI for model management
Custom Endpoint	Any OpenAI-compatible API	Self-hosted models, proxies

Local providers are ideal for privacy-sensitive work, offline environments, and cost savings (no per-token charges).

🔧 Setting Up Providers

To add a provider:

1. 1. Go to Settings → Providers
2. 2. Click on the provider you want to add
3. 3. Enter your API key (for cloud providers)
4. 4. Click "Validate" to test the connection
5. 5. Select which models you want to use
6. 6. Configure optional settings (base URL, max tokens, etc.)

For local providers, just install the software (Ollama/LM Studio) and Aura Work will auto-discover available models.

🔑 Credential Security

API keys are encrypted using the device-bound vault before storage. The vault uses:

• Windows — DPAPI (Data Protection API)
• macOS — Keychain
• Linux — Secret Service (GNOME Keyring / KWallet)

Credentials are never logged, never exposed to the agent, and never included in audit entries. The vault supports biometric unlock on supported platforms.

📊 Usage Tracking

Every task records detailed usage information:

• Input tokens — tokens sent to the model
• Output tokens — tokens received from the model
• Estimated cost — calculated from the pricing cache
• Model used — which provider and model handled the task

View usage statistics in the Dashboard or export them for billing. The audit log maintains a permanent record of all provider interactions.

🔄 Fallback & Retry

If a provider fails, Aura Work can automatically retry with an alternative:

1. 1. Primary provider fails (rate limit, timeout, error)
2. 2. System checks for fallback providers in your configuration
3. 3. If fallback exists, retries with the alternative provider
4. 4. If no fallback, notifies you and asks for manual intervention

Configure fallback providers in Settings → Providers → Fallback Chain.

💡 Cost Optimization Tips

• Use the cost-first routing policy for routine tasks
• Set up Ollama for development and testing (free)
• Use DeepSeek for code tasks (cheaper than OpenAI/Anthropic)
• Monitor usage in the Dashboard to identify expensive patterns
• Set token limits per task to prevent runaway costs

🎯 Model Selection Guide

• For complex coding — Claude Sonnet 4, GPT-4o, or DeepSeek R1 (advanced reasoning)
• For fast/simple tasks — Claude 3.5 Haiku, GPT-4o mini, or Gemini 2.0 Flash
• For very long context — Gemini 2.5 Pro (massive context window)
• For maximum privacy — any model via Ollama or LM Studio
• For lowest possible cost — DeepSeek V3 or a local model via Ollama

Cloud

• OpenAI — GPT-4o, GPT-4.5, o1, o3. Excellent at coding and analysis
• Anthropic — Claude Opus, Sonnet, Haiku. Long context and computer vision
• Google Gemini — Gemini 2.5 Pro, Flash. Multimodal and lightweight models
• DeepSeek — DeepSeek-V3, R1. Strong performance at low cost
• xAI Grok — Grok-3. Cutting-edge models from xAI

Local

• Ollama — Llama, Mistral, Gemma, Qwen. Complete privacy
• LM Studio — a graphical interface for running models locally
• LocalAI — a local OpenAI-compatible API