Hex
Query results live in session memory, not on disk — reopening a saved project restores the code, not the computed state. A genuinely worth-knowing limitation for a notebook tool built around real-time collaboration.
What is Hex?
Hex, founded in 2019 by former Palantir employees Barry McCardel, Caitlin Colgrove, and Glen Takahashi, is a collaborative data workspace sometimes nicknamed "the Figma for data teams" — a fitting comparison, since it turns the traditionally solo experience of a Jupyter notebook into a genuinely real-time collaborative product. Multiple team members can join the same notebook simultaneously — one writing SQL to slice a dataset, another writing Python for deeper statistical analysis, both annotating findings inline as they work — creating a real, shared investigation record rather than passing screenshots and static exports back and forth. What genuinely distinguishes Hex is the depth of its AI layer: the Notebook Agent autonomously generates and edits analysis logic with real awareness of your warehouse schemas and project history, Threads provides a natural-language query interface for non-coders to ask data questions directly, and Magic AI handles code generation, debugging, and statistical summaries across SQL, Python, and R. Companies like Notion and Stubhub use Hex specifically to get trusted, fast answers and build interactive dashboards, and once an analysis is built, publishing it as a polished, shareable stakeholder app — complete with adjustable filters and date ranges but no visible code — takes one click, backed by real version control and scheduled runs that make it genuinely production-grade rather than purely experimental.
The genuinely important, specific technical detail worth knowing before relying on Hex for long-running analysis: query results live in session memory, not on disk, meaning opening a previously saved project restores your code but does NOT restore the computed state — you'll need to re-run the notebook to see results again, a real, worth-understanding architectural choice rather than a bug. It's also worth knowing that the Notebook Agent specifically requires an Editor role or above on a paid plan; Viewer accounts cannot run it at all. On pricing specifically, Hex takes a genuinely different, arguably fairer approach than some competitors in this category: per-editor pricing runs $36 to $75 a month depending on tier, but viewer seats are billed separately and don't compound the way editor seats do — a real, meaningful contrast to a platform like Mode Analytics, also covered elsewhere in this directory, where both editors and viewers count identically toward your total seat cost. It's fair to flag honest, specific scope limitations too: Hex is genuinely overkill for simple spreadsheet tasks (a tool like Julius AI, also covered elsewhere in this directory, is better suited to quick CSV analysis), it requires real database access to deliver full value, its Python environment carries some package limitations compared to local Jupyter, and Magic AI's SQL accuracy depends heavily on how well-documented your schema actually is.
The session-memory architecture and Notebook Agent access requirements are drawn directly from an independent 2026 hands-on guide (VSLZ AI). The separate-viewer-seat pricing model and honest scope limitations are drawn from two independent reviews (Agents AI and Build Fast with AI). Company founding details and funding figures are drawn from Agents AI's independent company profile.
Key features
Notebook Agent
Autonomously writes and edits analysis logic with awareness of schema and project history.
Threads
Natural-language query interface letting non-coders ask data questions directly.
Real-time collaborative notebooks
Multiple people writing SQL, Python, and annotations simultaneously in one project.
One-click published apps
Turn a notebook into a polished, interactive stakeholder dashboard instantly.
Semantic model governance
Endorse trusted data definitions so AI agents give consistent, verified answers.
Scheduled runs
Re-execute a notebook daily or hourly and push fresh results automatically.
Pricing
Free
- Published apps capped (varies by source, roughly 2-5)
- Genuine full access to core notebook features for individual evaluation
- Notebook Agent access still requires a paid plan to run
Individual / Team
- Team adds unlimited published apps, shared components, scheduling
- Per-editor cost compounds for larger teams, but viewers don't
- Notebook Agent requires Editor role or above specifically
Enterprise
- Best for organizations needing centralized data governance at scale
- Zero data retention agreements with AI providers included
- Confirm current terms directly given frequent feature updates
Query results are not persisted to disk between sessions — plan for re-running notebooks rather than expecting instant reload of previous computed results. Prices reflect Hex's published pricing as of July 2026.
Available models
Integrations & platforms
Pros, cons & best for
Pros
- Genuinely deep, schema-aware AI integration beyond simple code autocomplete
- Real-time, multi-person collaborative notebooks are a genuine, tangible workflow upgrade
- Viewer seats billed separately from editors, a fairer pricing model than some competitors
Cons
- Query results live only in session memory, not restored automatically on reopen
- Genuine overkill for simple spreadsheet or CSV-only analysis tasks
- Requires real database access and well-documented schema for full AI accuracy
Best for
- Data teams working directly with cloud data warehouses like Snowflake
- Collaborative investigations, anomaly diagnosis, and ad-hoc data quality work
- Not the pick for simple spreadsheet analysis without a warehouse behind it
Take a look inside
Alternatives
For simple spreadsheet analysis or SQL-first collaborative reporting instead:
Our verdict
Hex's genuine strength is real, deep AI integration built specifically for data teams — the Notebook Agent's schema-and-history awareness, combined with genuinely real-time collaborative editing, reflects a thoughtfully built product that earns its "Figma for data teams" nickname honestly. Its fairer pricing model, billing viewer seats separately from editors, is a real, meaningful advantage over competitors that count every viewer as a full seat. The honest, specific technical detail worth understanding before relying on Hex for long-running work: query results live only in session memory, not on disk, so reopening a saved project restores your code but not your previously computed results — a real architectural choice worth planning around rather than assuming instant reload. Combined with honest, fair scope limitations (genuine overkill for simple CSV work, real database access required for full value), Hex remains a strong, well-differentiated, and well-funded choice specifically for data teams doing genuine collaborative analysis on cloud data warehouses.
FAQ
Does reopening a Hex project restore my previous results?
Not automatically — query results live in session memory, not on disk, so reopening a saved project restores your code but you'll need to re-run the notebook to see computed results again.
Does Hex charge for dashboard viewers the same as editors?
No, genuinely not — viewer seats are billed separately from editor seats, a fairer pricing model than some competitors in this category where viewers count identically toward total seat cost.
Can anyone use Hex's Notebook Agent?
No, it requires an Editor role or above on a paid plan specifically — Viewer accounts cannot run the Notebook Agent at all.
Is Hex good for simple spreadsheet or CSV analysis?
Not really its strength — Hex is genuinely overkill for simple spreadsheet tasks, and a tool like Julius AI or ChatGPT is generally better suited to quick, conversational CSV analysis specifically.
Does Hex require a data warehouse to be useful?
To get full value, yes — Hex requires real database access, and it's genuinely less useful without a connected data warehouse behind it.
Is my data used to train Hex's AI models?
No, Hex's AI providers operate under zero data retention agreements, meaning customer data is not used for model training — a real, meaningful trust signal for data-sensitive organizations.