OpenAI has launched a Data agent in ChatGPT Work that can investigate connected company data, build interactive dashboards and prepare approved follow-up actions.
RELATED BUYER PROFILEReview the Data agent access, cost and test-pending profilePricing · access · strengths · limitations →CONTINUE THE DECISIONBuild a controlled AI stack for a finance teamEvidence · workflow · next action →CONTINUE THE DECISIONRun the connected vendors through due diligenceEvidence · workflow · next action →CONTINUE THE DECISIONCalculate complete cost per accepted analysisEvidence · workflow · next action →What you need to know
- The Data agent can investigate connected business data, explain changes and create interactive dashboards in ChatGPT Work and Codex
- It depends on both the Data plugin and separately authorised data-source plugins; installation does not grant access to the underlying data
- OpenAI has not announced a standalone Data agent price or free trial, so workspace seats and connected vendor licences shape total cost
The deployment boundary at a glance
What launched on 10 September
OpenAI has released a Data agent inside ChatGPT Work. It can connect to approved warehouses and business systems, investigate a question in natural language, inspect changes over time and turn the analysis into an interactive dashboard or report. The launch page names sources including Amazon Redshift, ClickHouse, Databricks, Google BigQuery, MongoDB and Snowflake, plus documents from Google Drive and SharePoint when those connections are available.
The semantic layer is the real accuracy control
The agent can use an organisation's metric definitions, calculations and data relationships from trusted semantic layers, documentation and dashboards. That context matters more than a polished chart. A buyer should define revenue, active customer, churn and every other consequential metric before asking the agent to diagnose movement. If its result differs from an existing report, OpenAI's own guidance says to compare the source, period, filters and metric definition.
Installation and data access are separate
A workspace administrator can make the Data plugin available or pre-install it for selected roles or groups. The required data-source plugin and included app still need their own setup, authorisation and workspace access. Queries use the connected account's existing table, row and column restrictions. This reduces accidental privilege expansion, but teams still need to test whether every connector, action and approval behaves as expected in their environment.
Dashboards can leave the analysis conversation
The Data agent can create, edit, share and refresh dashboards and can work with BI tools including Omni, Oracle BI, Power BI, Sigma, Tableau and ThoughtSpot. OpenAI's help guide adds an important publication warning: data used in an analysis is copied into a published Site. The publisher must therefore review the destination, audience and included data before sharing; source-system permissions do not automatically make a published artefact safe for a wider audience.
Price is a workspace decision, not a free add-on claim
OpenAI does not list a separate price or free trial for the Data agent. Its business pricing page currently includes ChatGPT Work and plugins in Business plans, with Standard seats listed at $20 per month when billed annually or $25 monthly, and Premium seats at $100 annually or $125 monthly. Enterprise pricing is custom. Warehouse, BI and enrichment products may carry separate contracts and usage costs, so total cost should include seats, connector licences, data compute, setup and human review.
HubAI buyer verdict
The Data agent is a strong candidate for teams that already maintain governed data sources and shared business definitions. It is not a substitute for that foundation. Pilot one recurring question against an analyst-approved baseline, record wrong joins, disputed definitions, correction time and refresh behaviour, then compare cost per accepted decision brief with the existing BI workflow. HubAI is withholding a score until that same-question test is complete.
HUBAI VIEWThe buying decision is not whether it can draw a chart: it is whether the semantic layer, account permissions and publication boundary make the result trustworthy and safe to share.
Buyer decision signal: New enterprise product · data and BI
What to verify next
1Confirm Data is available in the required ChatGPT Work or Codex workspace
2Authorise the Data plugin and each source plugin separately
3Document metric definitions, joins, time zones and approved source hierarchy
4Test table, row and column restrictions with representative user roles
5Compare results with an analyst-approved report and record correction time
6Review copied data and audience before publishing any dashboard or Site
7Calculate seats, connected-vendor licences, compute and review as one complete cost
8Keep consequential Slack, email or connected-tool actions behind human approval
Read the evidence
Capabilities, availability and prices can change. HubAI keeps analysis separate from the underlying official material.
01OpenAI: Now everyone can put data to workOpen source ↗02OpenAI Help: Using the Data plugin in ChatGPT Work and CodexOpen source ↗03OpenAI: ChatGPT Work for data teamsOpen source ↗04OpenAI business pricingOpen source ↗
Code AI
Image AI
Safety
Safety