Documentation & API Reference

Quickstart Installation

MinTok can be installed across all major platforms using your preferred package manager, standalone script, or dedicated desktop studio:

curl -fsSL https://mintok.adstim.net/install | bash

npm install -g mintok
# or run without installing: npx mintok --version

brew install mintok/tap/mintok
# or direct formula: brew install https://mintok.adstim.net/formula/mintok.rb

Authentication

Authenticate the CLI using your API key from the Developer Dashboard:

mintok login --key mtk_live_9f81a7b...

Keys can also be passed via the MINTOK_API_KEY environment variable.

Available Models & Pricing

All models are served via dedicated Serverless GPU pools with automatic scale-to-zero economics:

OpenAI-Compatible Chat Completions

Directly target MinTok endpoints from the standard OpenAI Python or TypeScript SDK:

from openai import OpenAI

client = OpenAI(
    base_url="https://mintok.adstim.net/v1",
    api_key="mtk_live_your_key_here"
)

response = client.chat.completions.create(
    model="mintok-1-pro",
    messages=[
        {"role": "system", "content": "You are an autonomous senior engineer."},
        {"role": "user", "content": "Refactor the database queries in auth.py to eliminate N+1 scans."}
    ],
    temperature=0.2
)

print(response.choices[0].message.content)

Streaming & Provenance Headers

Every completion response streams cryptographic metadata proving avoided spend:

AST Context Compiler Mechanics

MinTok compiles agent conversation history before GPU dispatch:

• AST Slicing: Unreferenced declarations and dead code are pruned from injected context blocks.
• Output Virtualization: Repetitive bash outputs and test traces are hashed and compressed to minimal failure summaries.
• Deterministic Stopping: Guard checks terminate runaway retries early, saving up to 40,000 tokens on failing tasks.