Hex AI review (Oct 2026): the Hex Agent, Threads, and Generative Apps for data questions

What this Hex AI page covers

This page is for the person who gets the “can you pull this number?” messages — an analytics lead, a finance or RevOps analyst, or an ops manager at a company whose data already sits in a warehouse such as Snowflake, BigQuery, Databricks, or ClickHouse. The question it answers: if we turn on Hex’s AI features, what can non-analysts safely do on their own, and what still needs a person who knows the data?

Hex’s AI work happens in a few places. Threads is the chat-style space where someone asks a question and the agent writes and runs the queries. The notebook agent edits SQL and Python cells for people who build. Generative Apps turn a prompt into a shareable data app. Context Studio is where admins curate what the agent should trust and watch how it is used. The thread that ties them together is endorsement: Hex lets you mark tables, semantic models, and projects as Endorsed, and its agents prioritize those first.

Everything below comes from Hex’s public pricing page, docs, and changelog as we read them in early October 2026. Feature names move quickly in this category, so check your own workspace before you promise a button to your team.

Major features (with how someone uses them)

Examples at a glance: each feature has a concrete example from a real job plus a short Try this you can run this week.

1. Threads with Endorsed Mode (primary)

What it is: Threads is a conversational space: you type a question, and the Hex Agent searches the connections you are allowed to use, writes SQL, runs it, and answers with tables and charts. Hex’s docs say the agent heavily prioritizes semantic models and endorsed data. With Endorsed Mode on, it only uses endorsed assets; by default, people with the Explorer role can only use Threads in Endorsed Mode, though admins can change that. Every Thread is backed by an unlisted project, so an Editor can flip to the Notebook view and read the exact logic the agent used.

Threads also turns up outside Hex: a Slack integration returns a Threads link with the agent’s answer, and MCP clients such as Claude or Cursor can start a Thread. Answers show the agent’s tool calls and thinking steps, collapsed after it finishes, so you can expand them and see which tables it touched.

Concrete example (ops manager) An ops manager asks, “How many orders shipped late last week by warehouse?” Endorsed Mode is on, so the agent uses the endorsed fct_shipments model instead of a half-finished staging table. The answer comes back as a bar chart. Before forwarding it, the manager opens the Notebook view and sees the late-shipment rule the data team defined — promised date before actual ship date — not a guess.

Try this

Primary workflow:

Turn on Endorsed Mode. Ask one recurring business question in a Thread. Open the Notebook view and confirm every table used is one you would have picked. If one isn’t, endorse the right one instead of re-asking.

2. Notebook agent with approval modes

What it is: For people who build, the agent works inside a regular Hex notebook — adding SQL and Python cells, fixing errors, and building charts. Since mid-September 2026, the prompt bar has two modes. Ask for approval pauses before destructive actions such as deleting a cell or publishing. Skip approvals lets the agent work straight through, and you can still undo the changes from its last turn. Hex says anything that widens access — granting OAuth, reconnecting an app, authorizing an MCP server — still stops for your approval in either mode.

Concrete example (finance analyst) A finance analyst has a month-end revenue notebook. She asks the agent, in Ask-for-approval mode, to add a cohort breakdown by signup month. It proposes two new SQL cells and a chart. She accepts the SQL, rejects a cell that would have overwritten her existing currency conversion, and reruns the notebook to make sure the totals still tie to the general ledger export.

Try this

Build check:

In a copy of one notebook, ask the agent to add one breakdown in Ask-for-approval mode. Accept only the cells you could explain to a colleague. Rerun and compare one total to the original.

3. Generative Apps and scheduled exports

What it is: Generative Apps (introduced May 2026) turn a prompt into an interactive data app. Since then Hex has added point-and-click chart controls in a Style panel (27 Aug 2026), direct text editing and targeted chart updates (10 Sep), and on 23 Sep both Chat with Generative Apps — viewers can ask follow-up questions in a Thread without leaving the app — and scheduled CSV exports alongside app notifications. Hex’s own note on exports is worth repeating: scheduled exports use the project’s underlying cell outputs, not the filters someone applied in the app, so check those outputs contain only what the recipients should see.

Concrete example (RevOps lead) A RevOps lead prompts an app showing pipeline by stage and region, then uses the Style panel to fix the legend instead of spending another prompt on it. Before scheduling a Monday CSV to regional managers, he opens the cell outputs and sees they include every region and rep email. He adds a filter cell so each export only carries the columns managers need.

Try this

Before you schedule:

Build an app from one endorsed query. Before scheduling a CSV export, open the cell outputs and delete any column the recipients should not see. Then schedule it to yourself first.

6. Versus ThoughtSpot, Power BI, and a chat assistant

What it is: Hex fits teams where analysts already write SQL and Python and want business users to self-serve on top of that work. ThoughtSpot Spotter fits companies that want a search-first BI tool for business users on a governed model. Power BI Copilot fits Microsoft shops whose reports already live in Power BI. A general chat assistant such as Claude with artifacts is fine for a one-off pasted CSV, but it does not know your warehouse or your definitions.

Concrete example (boundary) A 40-person company keeps finance reports in Power BI because the CFO lives in Excel and Teams. The two-person data team uses Hex for ad-hoc analysis and a handful of apps. Nobody buys a third BI seat to “ask the data” — they point Threads at the same endorsed models the Power BI reports use and keep definitions in one place.

Try this

Home for the job:

List the six data questions people asked you last month. For each, write where it should be answered — Hex, your BI tool, or a one-off chat — and one sentence why. Do not add a tool you do not already pay for.

Pricing context (October 2026)

Hex publishes prices on hex.tech/pricing. When we checked in early October 2026, the page listed: Community free (with a Hex agent trial and small compute); Professional at $36 per Editor per month (Hex agent with standard credits, up to 5 published apps); Team at $75 per Editor per month with a 14-day free trial (extended credits, unlimited published apps, scheduled runs and alerts); and Enterprise at custom pricing (premium credits, Explorer seat add-on, audit logs, BYOK, single-tenant and HIPAA add-ons).

AI features use credits. Hex’s FAQ says Professional and Team include monthly per-seat credit grants that refresh monthly and do not roll over, with add-on credits available; Enterprise grants are set on the order form. The changelog also mentions spend limits and credit usage controls (July 2026) and a Fast mode for some models “at a higher credit cost” (September 2026). Medium compute is included on paid plans; larger compute profiles are billed per minute.

Uncertainty: Hex changes plan contents and credit rules often, and the price of a heavy agent month depends on credits, not just seats. Whether your plan bills monthly or annually, how many credits a typical Thread uses, and which add-ons you need are questions for Hex’s live pricing page and your account team before you budget.

When it’s useful (honest take)

Useful when your data already sits in a warehouse, someone on the team writes SQL, and you are willing to endorse the tables and models people should trust. Then Threads can take the “can you pull this?” load off the data team, and the Notebook view keeps every answer checkable.

Less useful when you have no warehouse and your data lives in a few spreadsheets — a spreadsheet tool with AI, or a chat assistant on a pasted file, is lighter. Less useful when nobody owns definitions: without endorsed data and guides, the agent will happily answer from whatever table looks closest. And if your company standard is already Power BI or Tableau for business users, Hex is more likely a tool for the data team than a replacement BI seat for everyone.

Related options (educational)

FAQ

What is Hex AI for on this page?

Plain-English questions on warehouse data (Threads), agent help inside notebooks, prompt-built data apps, and admin tools to curate what the agent trusts. Check which features your plan includes.

What does Endorsed Mode do?

It limits the agent to endorsed tables, semantic models, and projects. Hex’s docs say Explorers use Threads in Endorsed Mode by default, and admins can change that.

How much does Hex cost in October 2026?

Hex’s pricing page listed Community free, Professional $36 and Team $75 per Editor per month, and custom Enterprise when we checked in early October 2026. AI use draws on credits. Confirm on hex.tech/pricing.

Should the agent run with approvals skipped?

Only on work you can undo and review. Skip approvals lets it work straight through; Hex says access-widening steps still pause. Publishing to others should stay a human decision.

Do we need a warehouse?

In practice, yes, for Hex to shine. For a single spreadsheet, a spreadsheet tool with AI or a chat assistant on a pasted file is lighter.

Optional next step (if seats overlap)

If you pay for a notebook tool, a BI tool, and a chat assistant that all answer the same “what was revenue last month?” question, that overlap is worth a look. Map the overlap with the $19 AI toolkit audit or the $9 AI toolkit spreadsheet. Email Audit $19 or Toolkit spreadsheet $9.

Checkout is live via Stripe. After you pay, we email delivery (spreadsheet) or intake next steps (audit).

Related: ThoughtSpot Spotter · Power BI Copilot · Tableau Pulse · Claude for finance · ops hub