Today: PostgreSQL + pgvector for document retrieval, schema-aware Text-to-SQL with validation, and multi-model workflows that can call cloud models.
TODAY: SOVEREIGN DATA CORE
Your Database Stays the Source of Truth
AI writes queries and drafts actions. Your systems compute the results, and your people approve the changes.
Document Retrieval
PostgreSQL + pgvector indexes policies and SOPs, and every answer cites its source.
Schema-Aware Text-to-SQL
Generated SQL is parsed and checked against your schema, with read-only access, before it runs.
Human-in-the-Loop Controls
Actions that write to your systems wait for a named approver, and every step is logged.
ROADMAP
Local Inference, Benchmarks and Scale
Local inference: Serve open-source models such as Llama and Qwen on a customer’s own GPU server with NVIDIA NIM and TensorRT-LLM, so questions and data stay on the customer’s network.
Benchmarks: Publish accuracy, latency and cost per thousand queries for local and cloud models on the same ERP question set.
Larger document collections: Evaluate NVIDIA cuVS for GPU-accelerated vector search where policy and document collections grow large.