Skip to main content
Averta screens requests and responses regardless of the underlying AI provider. You specify the provider metadata when creating a decision so Averta can apply the right analysis pipeline.

​
Provider list

ProviderOperationsModels
OpenAIchat, completion, embeddingGPT-4o, GPT-4, GPT-3.5, o1, o3
AnthropicchatClaude 4.6, Claude 4.5, Claude 3.5
Googlechat, completionGemini 2.5, Gemini 2.0
MistralchatMistral Large, Medium, Small
Coherechat, completionCommand R+, Command R
CustomAnyAny model identifier

​
Provider metadata

When creating a decision, pass the provider block: 
const decision = await averta.decisions.create({
  provider: {
    name: "openai",        // provider identifier
    model: "gpt-4o",       // model name
    operation: "chat",     // operation type
  },
  messages: [...],
});
For custom or self-hosted models, use "custom" as the provider name and pass any string as the model: 
provider: {
  name: "custom",
  model: "my-fine-tuned-llama-3",
  operation: "chat",
}
Averta’s detection pipeline is model-agnostic — the provider metadata is used for analytics and checkpoint routing, not for the analysis itself.