Everything an AI agent needs to answer accurately: ontology, knowledge graph and decision traces.
<ul><li>Most agent failures are context failures: wrong SKU counts, stale prices, expired offers or unrecognised product nicknames.</li><li>A strong context layer improves accuracy, lowers cost and speeds up answers. The model can be anyone's; the context belongs to the business.</li><li>Evals are run by the agent harness. Their graded results are kept as decision traces, which become part of the context.</li></ul>