Aller au contenu
NeoXamAgents

Data manager · DataHub developer

Beta (V1)

Data onboarding & configuration Q&A

'How do I configure a Kafka feed?' — answered from documentation and live configuration objects, with deep links to the exact screens.

The DataHub assistant is the first conversational pilot of the Beta — Milestone β is the assistant live in product, grounded in documentation and configuration.

In DataHub

The daily reality

The work this replaces

Onboarding a new data flow means reverse-engineering documentation and existing feeds. The institutional knowledge lives in a few heads, so simple configuration questions queue behind busy experts and stall delivery.

How it works

What the agent actually does

datahub-assistant

  1. Ask inside DataHub

    The data manager asks in plain language — 'How do I configure a Kafka feed?' — in an assistant embedded in the product screen, not a separate portal.

  2. Grounded retrieval

    The agent retrieves from product documentation and live configuration objects (RAG), so answers reflect your instance — not a generic manual.

  3. Answers with deep links

    Responses link to the exact screens where the configuration happens, compressing the path from question to action.

  4. A governed conversation

    Sessions, platform-managed memory and streamed answers — declared in the agent's descriptor and traced like every other run.

The outcome

Outcome

Onboarding of new data flows compresses: configuration questions stop queuing behind the few people who know.

Governed by design

Conversational mode, sessions and live product connectors are Beta (V1) capabilities; the assistant is declared, eval-gated and traced like every platform agent.

Governance & evals

FAQ

Questions teams ask about this

When can we use it?

The DataHub assistant is the first pilot of the Beta (V1) — Milestone β is the assistant live in product. The Early Adopter Program is the way into that cohort.

How are answers kept accurate?

By grounding: retrieval over product documentation and live configuration objects, with deep links so users verify at the source. Like every agent, the assistant carries a versioned eval dataset and a baseline score that can only go up.

What does the assistant remember?

Memory is a declared policy — none, a sliding window, or a rolling summary — applied by the platform, never improvised by the agent. Sessions keep an archived history and survive infrastructure restarts.

Go further

  • Weeks of client-deployment parametrization and business-rule authoring compressed into days — with every draft reviewed by experts.

  • Write 'flag NAV deviations above 2%' — the generator returns a syntactically valid business rule as a typed object DataHub consumes directly.

  • Data-quality remediation

    In product · unified at GA

    Data-quality errors investigated in the context of the feeds, rules and configuration involved — so remediation becomes review-and-apply.

From the glossary

Join the Early Adopter Program

General availability lands in Q3 2026. The Early Adopter Program is open now — a limited cohort, a one-year platform trial, and three workshop streams: Business ROI, Compliance, and Operational fit.