Hex

🏢 Who's behind it

Hex is a cloud-based collaborative data platform — hex.tech — founded in 2019 by a group of ex-Palantir engineers. The founders cut their teeth building data products at Palantir, and they carried over a habit from those days: writing code and conclusions in the same notebook, then wrapping it in collaboration and publishing for everyone else.

The bet behind the company is that data teams don't lack tools, they lack one place where SQL, Python, and dashboards live on the same page. Hex bills itself as the modern replacement for Jupyter plus a standalone BI tool — not a pure code notebook, not a drag-and-drop dashboard, but the middle ground. In April 2025 it bought Hashboard, a BI visualization and semantic-modeling company, and the same year it closed a $70 million Series C with Snowflake, a16z, Sequoia, and Amplify participating. Snowflake backing it is a signal that Hex has a real seat in the modern data stack.

📅 When it launched

Hex went live in 2019. By then Jupyter had been popular for nearly a decade, but collaboration and publishing were still its weak spots — Hex aimed straight at that gap.

The milestones since: April 2025, the Hashboard acquisition to fill out BI and semantic modeling; later in 2025, the $70 million Series C. By 2024 revenue was around $19.8 million and more than 1,500 teams were on the platform. The arc reads as establish the collaborative notebook first, then buy and raise your way upmarket.

🧩 What it does

In one sentence: it lets you mix SQL, Python, R, and no-code cells in a notebook-style interface, then publish the result as an interactive data app anyone can open. Broken out:

• Multi-language notebooks: SQL, Python, R, and no-code cells side by side, so you query, compute, chart, and write conclusions in one page.
• Hex Magic: natural-language SQL or Python generation that produces reviewable drafts instead of black-box results.
• Notebook Agent: a Claude-powered agent that plans and runs multi-step analyses — querying tables, building charts, writing conclusions.
• Threads agent: natural-language data Q&A aimed at business users.
• Semantic model agent and dbt semantic layer: one source of truth for metric definitions, so everyone quotes the same number.
• One-click interactive data apps: turn an analysis into a dashboard with filters and dropdowns your colleagues can actually use.
• Real-time collaboration: Google Docs-style multi-user editing, version history, and inline comments.
• Scheduled runs and notifications: set a job, get the result by email or Slack.
• Embedding: drop results into your own product.
• Connectors: Snowflake, BigQuery, Databricks, Redshift, Postgres, MySQL, and other major sources.

✨ The highlights

• SQL + Python + dashboards in one environment: no more hopping between Jupyter, a BI tool, and a dashboard — you stop paying the context-switch tax.
• Data science is a first-class citizen: pandas, scikit-learn, TensorFlow, and D3.js run directly, so modeling and presenting stay on the same line.
• Collaboration feels like Google Docs: live multi-user editing, comments, and version history — the thing notebooks have always lacked.
• AI that goes from assist to agent: Magic generates code, Notebook Agent runs a whole analysis; the level of automation ramps up in steps.

🤖 The AI, honestly

Hex's AI isn't a chat box bolted on; it's layered:

• Hex Magic: describe what you want in a cell and it drafts SQL or Python. The important part is that it hands you a draft to review before you run it, not a final answer.
• Notebook Agent: a Claude-driven agent. Give it a goal and it decomposes the task, queries tables, writes code, charts it, and lands on a conclusion. Built for open-ended jobs like 'why did conversion drop this month?'
• Threads agent: a Q&A entry point for business users, grounded by the semantic layer so numbers stay consistent.
• Semantic model agent: works with the dbt semantic layer to unify metric definitions, cutting down on 'same number, two different values.'

The smart design choice is that generated code is always reviewable and every agent step lands visibly in the notebook — nothing is a one-shot black box.

🎯 Real-world uses

• Analytics teams: analysts pull with SQL, data scientists model in Python, and both hand off work in the same notebook.
• Product and growth teams: retention, conversion, and channel metrics become interactive apps that business folks can self-serve instead of queuing for the analyst.
• Machine learning teams: feature engineering and modeling with pandas, scikit-learn, TensorFlow, plus custom D3.js charts, all without leaving one page.
• Executive reporting: scheduled jobs refresh weekly and push to Slack, so the numbers are already done by the time anyone asks.

🛠️ How to use it

• Sign up and start on the free Community plan.
• Connect a data source — Snowflake, BigQuery, Postgres, MySQL — and enter credentials to read tables.
• Create a notebook, drop in a SQL cell, and write a query; add a Python cell when you need to process results.
• Use Hex Magic to describe the request in plain language, review the generated code, then run it.
• For multi-step work, hand Notebook Agent a goal and let it plan and execute.
• Show results in chart or table cells, with titles and a written conclusion.
• Hit Publish to turn it into a data app, set filters and permissions, and share the link.
• Add a scheduled run with email or Slack notifications for recurring refreshes.

For your first pass, the minimal loop is connect a warehouse, pull with SQL, have Magic write a Python transform, publish as an app. Add agents and collaboration once that path feels familiar.

💡 Tips

• Read Magic's SQL before you run it: it produces drafts, and joins or window functions deserve a once-over before you trust the result.
• Put shared definitions in the semantic layer: use the Semantic model agent or dbt so the team stops arguing about metrics.
• Ask Notebook Agent to show its plan first: for 'why' questions, confirm the steps before it executes, so it doesn't run off course.
• Don't push large data through: millions of rows and chained queries lag — aggregate in the warehouse first, then bring it into Hex.
• Watch compute spend: compute is metered (roughly $0.32 to $4.06 per hour), so estimate before you schedule long-running jobs.
• Lock down filters and permissions before publishing: an app shared with business users is only as safe as its access controls.

📈 Real examples

Hex reports more than 1,500 teams on the platform, about $19.8 million in 2024 revenue, and a G2 score around 4.8 out of 5. Underneath those numbers is the mixed team reality: data scientists, analysts, and business stakeholders sharing one analysis without everyone needing to learn Jupyter.

A typical pattern is the product team's weekly dashboard — an analyst builds retention, conversion, and channel views into an app, and product managers click through it daily instead of filing tickets. Data science teams use it for exploratory work and model prototypes, then demo the whole thing on one page rather than exporting notebooks to slides.

📱 Platforms

Hex is a cloud web app — open hex.tech in a browser and you're in, with no client to install. Major desktop browsers all work. Because SQL and Python run on cloud compute, your local machine barely matters; any computer with a browser and a connection gets you there.

There's no desktop app in the traditional sense, and everything lives in the cloud, so switching machines or going remote doesn't interrupt your work.

🌐 Languages

'Languages' here works on two levels. One is the notebook's programming languages: SQL, Python, and R, plus no-code cells, all mixable in a single notebook. The other is the product UI, which is English-first with no full multilingual interface — worth knowing before a non-English team commits.

On the data side, connectors cover Snowflake, BigQuery, Databricks, Redshift, Postgres, MySQL, and most of the major warehouses and databases.

💰 Pricing

Pricing is per-seat plus metered compute, across four tiers:

• Community (free): up to 5 notebooks/projects and 5 apps, limited AI credits, roughly 100 Hex Magic runs a month.
• Professional, $36/editor/month: unlimited notebooks/projects, unlimited Magic, Notebook Agent, 30-day version history, and scheduled runs.
• Team, $75/editor/month: adds Threads agent, Semantic model agent, an Explorer role, review and approval, user groups, GitHub/GitLab/Airflow/Dagster/dbt integrations, and unlimited version history.
• Enterprise (custom): SSO/Okta/OIDC, single-tenant, HIPAA, audit logs, Snowflake Marketplace, and embedded analytics.

Compute is an add-on: about $0.32/hour on a 16GB tier up to $4.06/hour on an A10G GPU tier. The catch — per-minute metering — makes budgeting fuzzy, and the cons section goes into that.

👥 Who it's for

• Mixed data science and analyst teams: people who need SQL and Python side by side and still want a dashboard at the end.
• ML-heavy teams: pandas, scikit-learn, TensorFlow, and D3.js run natively, so modeling and presentation aren't split.
• Teams that need live collaboration: multi-user editing, comments, and version history as a hard requirement.
• Teams wanting business self-service: publish analysis as apps and let business users click around.

The reverse case: pure business users who don't write SQL or Python won't get much out of it, and a drag-and-drop BI tool might serve them better. Budget-sensitive small teams should also weigh the enterprise pricing.

👍 Pros

• One environment for SQL + Python + dashboards: low switching cost, and results are actually shippable.
• Strong data science/ML workflow: mainstream libraries run directly, with D3.js for custom charts, all the way from modeling to presentation.
• Excellent real-time collaboration: Google Docs-style editing and comments beat traditional notebooks by a mile.
• Fits mixed teams: coders and non-coders can work on the same page.
• Layered AI: from Magic to agents, with reviewable code and visible steps.

⚠️ Cons and limits

• Advanced features assume SQL/Python: without code, half of Hex's value sits out of reach.
• Large data lags: millions of rows and chained queries slow things down; you'll want to aggregate in the warehouse first.
• Per-minute metering is hard to predict: compute is pay-as-you-go, and heavy batch jobs can surprise the bill.
• The AI can get preachy: generated output often comes wrapped in explanations and caveats that slow you down.
• Overlapping features confuse the UI: Magic, Notebook Agent, and Threads have blurry boundaries, and newcomers get lost.
• Enterprise pricing is steep: Team starts at $75/editor/month, and it adds up fast at scale.

❓ FAQ

Is there a free plan?

Yes — Community is free, capped at 5 notebooks/projects and 5 apps, with limited AI credits and roughly 100 Magic runs a month.

How is Hex different from Jupyter?

Hex takes Jupyter's notebook idea to the cloud and adds live collaboration, version history, one-click app publishing, and AI. Jupyter is more local, individual, and open source.

Can I use Hex without code?

You can ask questions through the Threads agent or open apps others have published, but getting full value means writing SQL or Python.

What is Notebook Agent?

A Claude-driven agent that plans and runs multi-step analyses — querying tables, building charts, writing conclusions — and lets you review its plan before it executes.

How is compute billed?

Compute is metered as an add-on, about $0.32/hour (16GB) to $4.06/hour (A10G GPU), billed by the minute.

Which data sources are supported?

Snowflake, BigQuery, Databricks, Redshift, Postgres, MySQL, and other major warehouses and databases.

Can I embed Hex in my product?

Yes, on the Enterprise plan, which includes embedded analytics.

🔄 Hex vs. the alternatives

• Jupyter Notebook: free, open source, local — but no collaboration or publishing. Hex wins on cloud collaboration, app publishing, and AI, at the cost of a subscription and cloud lock-in.
• Deepnote: both do cloud notebooks; Deepnote leans developer-friendly and lightweight, while Hex pushes further into enterprise with heavier AI agents and a semantic layer.
• Mode: also SQL-first analytics, but Hex's Python/ML and AI agent capabilities stand out more.
• Databricks: a heavy lakehouse platform for large-scale data processing; Hex fits teams with smaller data that want fast collaboration and app publishing.
• Tableau / Power BI: drag-and-drop BI, friendly to business users; Hex leans toward code and ML workflows, which analysts and data scientists prefer.