Hex

Hex is a data workspace built around one idea: the notebook you explore in and the app your stakeholders read should be the same object. You write SQL, Python, and no-code cells in a multiplayer notebook, then publish it as an interactive app without rebuilding anything. For analysts, that removes the usual second job of porting a finished analysis into a BI tool.

The AI layer has been renamed and split up. Hex's documentation says its AI features "were formerly referred to as Hex Magic in product"; today they are distinct, named agents with different jobs and different audiences.

Highlights

• Notebook Agent. A natural-language assistant that generates and edits Python, SQL, Markdown, Pivot, and Chart cells with your project context and warehouse schema in view, and understands the upstream and downstream dependencies in a flow well enough to debug it.
• Threads. Conversational analytics for people who will never open the notebook, grounded in curated, semantically modeled data. Hex surfaces it where the questions already get asked, including Slack and coding agents.
• Context Studio. The control plane for what the agents know: semantic models, workspace rules, endorsed tables, AI guides, and interaction-level observability so you can see which context produced which answer.
• Semantic model sync, not lock-in. Context Studio syncs semantic models from dbt MetricFlow, Cube, and Snowflake Semantic Views, and also draws on Git repositories, Notion, Linear, and any MCP server. Hex's own position is that semantic models are "inherently rigid and time intensive to maintain," so build them where a locked definition matters and skip them elsewhere.
• More agents around the edges. A Modeling Agent helps build and refine semantic models, Chat With App Agent lets app consumers interrogate a published project, and Generative Apps (beta) builds an app from a description.
• Hex CLI. If you would rather drive the work from Claude Code or Cursor, the CLI lets you bring your own agent instead of using Hex's.

In an analyst's workflow

The pattern that pays off is doing the context work once, then letting Threads absorb the repeat questions. Endorse the tables that are safe, write a guide for the metric everyone gets wrong, and only build a semantic model where a definition has to be locked:

# In the notebook, with the Notebook Agent:
Build a weekly cohort retention view from fct_subscription_events.
Use the endorsed `dim_accounts` table for plan tier. Show the SQL cell
and a chart cell, and explain which joins could double-count.

Then publish it as an app, and point the people who used to ask you for it at Threads. The text-to-SQL with Claude guide covers the same grounding problem when you are assembling the context yourself rather than in a vendor's studio.

Good to know

Hex is a hosted web platform. As of September 2026 it lists a free Community plan (Hex agent trial, any data source, small compute), Professional at $36 per Editor per month, Team at $75 per Editor per month, and custom Enterprise pricing that adds explorer seats, audit logs, OAuth database connections, OIDC SSO, BYOK, and embedded analytics. Agent usage is metered as credits, which scale by plan, so a heavy agent month costs more than a light one on the same seat count.

Choose Hex over a pure notebook when publishing and governance matter as much as the analysis. If you mainly want a fast, chat-first way to interrogate a file, Julius is lighter; if you want a notebook with a strong agent and a cheaper entry point, look at Deepnote. If your data and semantics already live in one platform, the built-in option there, such as Databricks Genie, is billed by consumption instead of per Editor, with user traffic free until January 31, 2027. The best AI tools for data analysts in 2026 roundup places all of them side by side.