DataHub says its tool lifts AI agent precision to near 100%

DataHub makes an open source platform for company data. It made DataHub Cloud Context available to all customers after a beta of five months. The tool gives AI agents the business facts about data, for example which table is the right one. DataHub says agents with no such context are right about 50% of the time. It says its method gives near 100% on primary metrics.

How it works

Three parts make a loop:

• Domain scoping: administrators put data, documents and test questions in groups, and give an agent only the groups that it uses.
• Evals: test questions that operate all the time and show when the precision of answers falls.
• Agent Feedback: agents use MCP, a standard connection to tools, to report missing or incorrect context.

The numbers

DataHub says its method gives near 100% on revenue metrics and near 90% on metrics that are not easy. Optiver, a trading firm, says that precision went from 42% to 74% on questions of researchers. Analyst Herb Blecher of EMA says 41% of firms report that incorrect AI output caused a bad step.

What is not known

The source gives no external test and no test data. All the precision numbers are from DataHub or from a customer.

Insight: Vendors fight to own the layer that tells agents what data means

In October 2026, AWS Context and Neo4j Context Engine came with no test results. Kong also has context management for agents. DataHub gives numbers, but they are from DataHub. This reads as a race to hold the context layer, and a vendor that holds it is hard to replace. Leaders must tell each vendor to show results on your data.