Hex
Collaborative SQL and Python notebook where an agent runs the analysis, wired into dbt, Cube or Snowflake semantic models so it answers against governed metric definitions.
Overview
Hex began as a better notebook. SQL and Python cells in one document, dataframes flowing between them, the output publishable as an app without exporting anything anywhere. The AI layer arrived in stages. Magic handled prompt-to-SQL and error explanation first. The Notebook Agent, which landed in late 2025, does the multi-step version: you give it a question and it runs the full loop of query, execute, chart and write-up, inside a project it can already read.
What separates it from a dozen text-to-SQL startups is the semantic model integration. Hex reads and syncs metric definitions from dbt MetricFlow, Cube and Snowflake Semantic Views, so when someone asks about revenue the agent applies the definition the data team already wrote rather than inferring a join across three tables and quietly getting it wrong. Context Studio sits above that for observing, testing and deploying the agent setups you expose to business users. That governed path is why data leads take Hex seriously in rooms where they dismiss chat-with-your-data demos out of hand.
The cost model needs watching. Professional runs $36 per editor per month, Team $75, and every paid seat carries a monthly credit grant that agent runs draw down, with top-ups sold separately. Add warehouse compute and a mid-size analytics team is spending real money. Hex is also deliberately not a data platform: no ingestion, no transformation scheduling, no catalog. Fine for a mature stack. Wrong first purchase for a team that has not built one.
Key Features
- ✓ Notebook Agent runs multi-step exploration on its own: writes the queries, executes cells, builds the charts and writes up the result, inside a project where it can already see your existing code and files
- ✓ Semantic model sync reads metric definitions from dbt MetricFlow, Cube or Snowflake Semantic Views, so the agent uses the definition your data team wrote instead of inferring joins
- ✓ Context Studio for observing, testing and deploying the agent configurations you expose to non-technical staff
- ✓ SQL and Python cells in one document with dataframes passing between them, and results cached across runs
- ✓ Notebooks compile straight into shareable apps with input widgets, scheduling and permissioned publishing, no separate BI tool in the chain
- ✓ dbt metadata and freshness checks surface in the notebook, so you can tell whether the model you are querying actually ran today
Where it holds
- • The semantic layer wiring is the real differentiator. Text-to-SQL against raw warehouse tables fails on join logic. Text-to-SQL against a governed metric definition mostly does not.
- • One document goes from exploration to published app, which collapses the usual notebook-to-dashboard handoff that wastes a day per analysis.
- • The free Community tier supports real work rather than a demo, though the agent is trial-only there.
Where it breaks
- • Per-editor pricing at $36 and $75 a month sits on top of your warehouse bill, and agent credits run down faster than teams budget for.
- • It is an analytics workspace, not a data platform. No ingestion, no transformation scheduling, no catalog, so it assumes dbt and a warehouse already exist upstream.
- • Sigma and Omni are cheaper and better fits if what you actually need is spreadsheet-style BI for business users rather than Python.
My Take
Best fit is a data team that already writes Python and has grown tired of the notebook-to-dashboard handoff. The Notebook Agent, which shipped in late 2025, is one of the few text-to-SQL implementations worth trusting in production, and the reason is architectural rather than model quality: it reads semantic models from dbt MetricFlow, Cube or Snowflake Semantic Views instead of guessing at raw table joins. Pricing is where the friction shows up. $75 per editor per month on Team, plus credits that every agent run consumes, plus warehouse compute, adds up fast across eight analysts. Take the 14-day trial before committing to seats.
Quick Info
- Pricing:
- freemium
- Openness:
- Proprietary
- Starting at:
- Community free, with small compute, up to 5 published apps and trial-only agent access. Professional $36 per editor per month. Team $75 per editor per month, adding advanced agents, unlimited published apps, shared components and scheduling, with a 14-day trial. Enterprise is custom. All paid seats carry a monthly credit grant that agent runs draw down, and extra credits are purchasable. Warehouse compute is separate and billed by your warehouse vendor.
- Added:
- Aug 2026
- Updated:
- Aug 2026
Use Cases
Judge it on your own work
The notes above say where Hex holds and where it breaks. The fastest check is your own workload.
Visit website ↗Alternatives to Hex
Databricks Mosaic AI
paidThe AI layer inside Databricks: Agent Bricks, model serving, vector search and MLflow, all resolving back to Unity Catalog governed data.
Gemini Notebook
freemiumGoogle's grounded research notebook, renamed from NotebookLM in July 2026. Upload sources, get answers with inline citations and run code against them in a per-notebook sandbox.