asi1-chat

ASI:One Chat

Overview

Query the ASI:One large language model — built by the Artificial Superintelligence (ASI) Alliance. OpenAI-compatible API with asi1 (default, recommended) and asi1-mini (faster, lighter) models.

When to Use

• User asks to "use ASI:One" or "query the ASI LLM"
• User asks to "use Fetch.ai's AI model"
• User wants an alternative to OpenAI/Anthropic for a specific query
• User asks about ASI Alliance AI capabilities

Prerequisites

• ASI_ONE_API_KEY environment variable set
  • Get one at: https://asi1.ai (sign up → API keys)
• Python 3.8+ with requests

Quick Steps

1. Simple query

python3 scripts/asi1_chat.py --prompt "Explain the ASI Alliance in 3 sentences"

2. With system prompt

python3 scripts/asi1_chat.py \
  --prompt "What are the benefits of decentralized AI?" \
  --system "You are a blockchain and AI expert. Be concise." \
  --model asi1

3. Streaming mode

python3 scripts/asi1_chat.py --prompt "Write a haiku about AI agents" --stream

4. Output

{
  "status": "success",
  "model": "asi1",
  "response": "The ASI Alliance is a collaboration between Fetch.ai, SingularityNET, and Ocean Protocol...",
  "usage": {
    "prompt_tokens": 12,
    "completion_tokens": 87,
    "total_tokens": 99
  }
}

API Details

• Base URL: https://api.asi1.ai/v1
• Endpoint: /chat/completions
• Models: asi1 (default, recommended), asi1-mini (faster, lighter)
• Format: OpenAI-compatible (same request/response format)
• Auth: Authorization: Bearer {ASI_ONE_API_KEY}

Also Works with OpenAI SDK

from openai import OpenAI

client = OpenAI(
    api_key=os.environ["ASI_ONE_API_KEY"],
    base_url="https://api.asi1.ai/v1"
)

response = client.chat.completions.create(
    model="asi1",
    messages=[{"role": "user", "content": "Hello!"}]
)
print(response.choices[0].message.content)

Edge Cases

• Rate limits: If 429 error, wait and retry
• Timeout: Large prompts may take 30–60s — increase --max-tokens or use --stream for real-time output. Use asi1-mini if you need lower latency.
• Streaming: Use --stream for real-time output on long responses
• Token limits: Context window varies by model — keep prompts reasonable