Nace.ai has officially launched Nace Document Intelligence (NDI) 1.0 and its companion decision engine, Drex, positioning itself as a specialized infrastructure layer for AI agents. The platform addresses a critical gap in agentic workflows: the need for low-latency, high-accuracy decision-making and verifiable document extraction without the overhead of massive LLMs. By offering a single API key for both services, Nace aims to simplify the integration of perception and decision logic for developers building autonomous systems.

Drex: Calibrated Probabilities for Agent Decisions

Drex is designed as a 'fast decision layer' that outputs calibrated probabilities for every option in a single forward pass. Unlike traditional LLMs that generate text, Drex focuses on routing tasks, classifying failures, and selecting tools with sub-second latency. Early adopters report significant improvements in efficiency, with one user noting that Drex handled 352 social media posts for spoiler detection in a single session while maintaining sub-second response times. The model is positioned as a more efficient alternative to using large language models for simple classification and guardrail tasks.

NDI: Verifiable Extraction Across 36 File Types

Nace Document Intelligence tackles the 'black box' problem of document processing by citing the source of every extracted value. The system supports 36 file types, including PDFs, scans, Word documents, and audio, using a combination of OCR, transcription, and NER models. Every extraction is grounded, providing the specific page and bounding box where the data was found. This level of transparency is crucial for enterprise workflows where auditability is mandatory, such as invoice processing where a $482,260 discrepancy was flagged by comparing invoice notes against purchase orders.

Flexible Deployment Models

Nace offers four distinct deployment tiers to accommodate varying security and compliance needs. The 'Our Cloud' option provides a shared GPU pool with strict tenant isolation for rapid prototyping. 'Dedicated Cloud' offers single-tenant compute with private links for regulated industries. For organizations requiring data sovereignty, 'Your Cloud' installs the platform directly into AWS or Azure VPCs, while 'On-Prem & Air-Gapped' delivers signed offline bundles for government and defense environments where no data can leave the network.

Key Takeaways

  • Drex provides calibrated probabilities for agent decisions, offering a faster and cheaper alternative to LLMs for routing and classification.
  • NDI supports 36 file types and grounds every extracted value to its specific page and bounding box for full auditability.
  • The platform offers four deployment models, ranging from shared cloud infrastructure to fully air-gapped on-premises installations.

The Bottom Line

Nace.ai is building the boring but essential plumbing for AI agents, moving beyond raw generation to reliable, verifiable decision-making and extraction. This is exactly the kind of specialized infrastructure that will make agentic workflows production-ready.