Sakana AI

Overview

Sakana AI is available as the sakana provider. It is useful for teams that want a learned model orchestrator: Fugu routes, delegates, and synthesizes work across a pool of frontier models behind a single OpenAI-compatible endpoint.

Fugu is the balanced, low-latency variant for everyday coding and chat. Fugu Ultra coordinates a deeper agent pool for maximum answer quality on complex, multi-step engineering and research tasks.

Setup

Create an API key from the Sakana AI get-started console, then confirm available model IDs in the Sakana AI models console.

export SAKANA_API_KEY="your-sakana-api-key"

autohand --provider sakana --model fugu-ultra-20260615

CLI configuration

{
  "provider": "sakana",
  "sakana": {
    "apiKey": "${SAKANA_API_KEY}",
    "baseUrl": "https://api.sakana.ai/v1",
    "model": "fugu-ultra-20260615"
  }
}

Models

Use /model to select a Fugu model or enter a custom model ID supported by your Sakana AI account. Check the Sakana AI models console for the current catalog.

/model sakana/fugu
/model sakana/fugu-ultra-20260615

Notes

• Fugu is not available in the EU/EEA at launch while Sakana AI works toward GDPR compliance.
• By default, Sakana AI may use responses to improve its models. You can opt out from the Sakana AI console.
• Fugu exposes orchestration token usage in prompt_tokens_details and completion_tokens_details, so you can see how many tokens the coordinator used behind the scenes.