AI Providers

Aetheris supports three AI providers: Google Gemini, OpenAI, and Anthropic Claude.

AI Providers

Configuration

Set the provider in .env:

AI_PROVIDER=gemini  # or openai, claude

⚠️ Security Warning: Never commit your .env file to version control. Ensure .env is listed in your .gitignore to prevent accidental credential leakage.

API Keys

Provider	Variable	Get Key
Gemini	`GEMINI_API_KEY`	Google AI Studio
OpenAI	`OPENAI_API_KEY`	OpenAI Platform
Claude	`ANTHROPIC_API_KEY`	Anthropic Console

Available Models

Gemini

Model	Notes
`gemini-2.5-pro`	Recommended, thinking mode (default)
`gemini-2.5-flash`	Fast, cost-effective
`gemini-2.5-flash-lite`	Ultra-fast, lowest cost
`gemini-2.0-flash`	Budget-friendly
`gemini-2.0-flash-lite`	Simple high-frequency tasks

Note: Gemini 1.5 and 1.0 models have been retired (404 error). Use gemini-2.0-flash for budget options or gemini-2.5-flash for better performance.

OpenAI

Model	Notes
`gpt-4o`	Recommended (default)
`gpt-4-turbo`	Fast
`gpt-4-turbo-preview`	Preview version
`gpt-4`	Stable
`o1-preview`	Reasoning model
`o1-mini`	Reasoning, faster
`gpt-3.5-turbo`	Budget-friendly

Claude

Model	Notes
`claude-3-5-sonnet-20241022`	Recommended (default)
`claude-3-opus-20240229`	Best quality
`claude-3-haiku-20240307`	Fast, cheap

Gemini Advanced Features

Available with Gemini models. These are configuration options set in AnalysisConfig or config.json:

Feature	Config Key	Default	Description
Structured Outputs	`enable_structured_outputs`	`true`	JSON validated via Pydantic
Thinking Mode	`enable_thinking`	`false`	Show reasoning process
Thinking Budget	`thinking_budget`	`1024`	Token budget for reasoning (min 128)
Code Execution	`enable_code_execution`	`false`	Python sandbox execution
Context Caching	`enable_context_caching`	`true`	~50% cost reduction
Cache TTL	`cache_ttl_seconds`	`3600`	Cache duration in seconds
Batch API	`enable_batch_mode`	`true`	~50% cost reduction
Google Search	`enable_google_search`	`false`	Web search grounding
Function Calling	`enable_function_calling`	`false`	Call Python functions

Example config.json:

{
  "enable_thinking": true,
  "thinking_budget": 2048,
  "enable_context_caching": true,
  "enable_google_search": false
}

Pricing

Prices change frequently. Consult official pricing pages for current rates:

Provider	Pricing Page
Gemini	Google AI Pricing
OpenAI	OpenAI Pricing
Claude	Anthropic Pricing

Cost Factors

Typical analysis costs depend on: - Project size: Number and length of files analyzed - Model choice: Premium models cost more than budget options - Token usage: Input (code) + Output (analysis) tokens

Budget-friendly options: Gemini Flash, GPT-3.5 Turbo, Claude Haiku Premium options: Gemini Pro, GPT-4o, Claude Sonnet/Opus

Recommendations

Use Case	Recommended
Budget-friendly	Gemini 2.0 Flash, GPT-3.5
Best value	Gemini 2.5 Pro, GPT-4o
Highest quality	Claude 3 Opus, GPT-4o
Code analysis	GPT-4o, Claude 3.5 Sonnet

Reasoning Display

To display model reasoning when available, set the show_reasoning config option:

{
  "show_reasoning": true
}

OpenAI Reasoning (o1-* models)

OpenAI o1-preview and o1-mini models include built-in reasoning. Set:

{
  "ai_model": "o1-preview",
  "show_reasoning": true
}

Gemini Thinking Mode

Gemini models (2.5+) support explicit thinking mode with configurable token budget:

{
  "enable_thinking": true,
  "thinking_budget": 2048
}

Note: Minimum thinking budget is 128 tokens. Budget is automatically capped at this minimum.