Configuration

Alumnium needs access to an AI model to work. The following models are supported:

Provider	Model
Anthropic	Claude 3 Haiku
Google	Gemini 2.0 Flash
OpenAI (default)	GPT-4o Mini
DeepSeek	DeepSeek V3
Meta	Llama 3.2 90B
Ollama	Mistral Small 3.1 24B

These models were chosen because they provide the best balance between intelligence, performance, and cost. They all behave roughly the same in Alumnium tests.

Trying out?

Google provides free-of-charge plan in many regions, which is convenient for experimenting. Alumnium automatically retries the requests when hitting rate limits on the free plans.

Anthropic

To use Anthropic as an AI provider in Alumnium:

1. Get the API key.
2. Export the following environment variables before running tests:

export ALUMNIUM_MODEL="anthropic"
export ANTHROPIC_API_KEY="sk-ant-..."

Google

To use Google AI Studio as an AI provider in Alumnium:

1. Get the API key.
2. Export the following environment variables before running tests:

export ALUMNIUM_MODEL="google"
export GOOGLE_API_KEY="..."

OpenAI

To use OpenAI as an AI provider in Alumnium:

1. Get the API key.
2. Export the following environment variables before running tests:

export ALUMNIUM_MODEL="openai"
export OPENAI_API_KEY="sk-proj-..."

DeepSeek

Caution

DeepSeek support is experimental and doesn’t work with vision checks. The current implementation works via the DeepSeek Platform, but we’re looking forward to extending it with Ollama, llama.cpp, etc.

To use DeepSeek as an AI provider in Alumnium:

1. Set up a DeepSeek Platform account.
2. Get the API key.
3. Export the following environment variable before running tests:

export ALUMNIUM_MODEL="deepseek"

Meta

Caution

Llama support is experimental and doesn’t work with vision checks. Its performance also highly depends on how you run it. The current implementation works via Amazon Bedrock, but we’re looking forward to extending it with Ollama, llama.cpp, etc.

To use Meta Llama as an AI provider in Alumnium:

1. Set up an Amazon Bedrock account.
2. Enable access to Llama 3.2 models.
3. Get the access key and secret.
4. Export the following environment variables before running tests:

export ALUMNIUM_MODEL="aws_meta"
export AWS_ACCESS_KEY="..."
export AWS_SECRET_KEY="..."

Ollama

Caution

Ollama support is experimental and performance depends on your hardware.

To use Ollama for a fully local model inference:

1. Download and install Ollama.
2. Download Mistrall Small 3.1 24B model:

ollama pull mistral-small3.1:24b

3. Export the following environment variable before running tests:

export ALUMNIUM_MODEL="ollama"

Read next to learn how to write tests!