Adaptensor Documentation¶

Private Document AI Built for Enterprise

Adaptensor is a private document AI platform that processes, indexes, and searches your documents using Google Cloud TPUs—delivering enterprise-grade AI compute at a fraction of GPU costs.

Core Technology: AdaptJax™¶

Our proprietary JAX-native embedding engine runs directly on Google Cloud TPUs, solving the "static shape barrier" that makes TPUs difficult to use with dynamic AI workloads.

Learn more about AdaptJax™

Quick Start¶

1. Install the SDK¶

pip install adaptensor

2. Initialize the Client¶

from adaptensor import Adaptensor

client = Adaptensor(api_key="your-api-key")

3. Upload and Search Documents¶

# Upload a document
doc = client.documents.upload("contract.pdf")

# Wait for processing
doc.wait_until_ready()

# Search your documents
results = client.search.query("What are the payment terms?")
for result in results:
    print(f"{result.score:.2f}: {result.text}")

View Full Quick Start Guide

Why Adaptensor?¶

🔒 Complete Privacy¶

Your documents never leave your infrastructure. Unlike cloud APIs, Adaptensor processes everything on-premise using your own TPU resources.

⚡ 10x Faster Than GPUs¶

Google's TPUs deliver superior throughput for AI inference. Our AdaptJax™ middleware makes them accessible without rewriting your code.

💰 25% Cost Savings¶

At $6.00/hr vs $8.00+/hr for A100 GPUs, you get better performance at lower cost.

🏢 Enterprise Ready¶

Built for legal, healthcare, finance, and aerospace sectors with strict compliance requirements (HIPAA, SOC2, GDPR ready).

Architecture: Compute Once, Query Forever¶

┌─────────────────────────────────────────────────────────┐ │ Your Documents (PDF, DOCX, TXT) │ └─────────────────────────┬───────────────────────────────┘ │ ▼ ┌─────────────────────────────────────────────────────────┐ │ INDEXING (One-Time TPU Compute) │ │ • Document chunking (semantic boundaries) │ │ • AdaptJax™ embedding generation │ │ • FAISS index construction │ └─────────────────────────┬───────────────────────────────┘ │ ▼ ┌─────────────────────────────────────────────────────────┐ │ QUERY (Instant, Repeated) │ │ • <20ms semantic search │ │ • Vector similarity matching │ │ • Optional LLM generation (RAG) │ └─────────────────────────────────────────────────────────┘

Platform Features¶

Documentation Sections¶

Patent Protection¶

Adaptensor technology is protected under U.S. Patent Application 63/930,053.