Hex

What it is

Hex is an analytics workspace built around a SQL-and-Python notebook that a data team uses for real analysis, with three AI layers on top. The notebook agent writes and fixes SQL, Python and charts inside a project. Threads is the self-serve layer: anyone with a seat asks a plain-language question, and the answer comes back as a chart and table backed by a full Hex project an analyst can open and audit. The modeling agent helps write and validate semantic model definitions, or you import existing ones from dbt MetricFlow, Cube or Snowflake Semantic Views. Threads is reachable outside Hex: mention @Hex in a Slack channel, or call Hex’s MCP server from Claude, ChatGPT, Cursor or Codex. Since September 15, 2026 those chat clients can also create and update Hex projects, not only read them.

2026 shipped the governance pieces that make non-analyst access defensible. Context Studio (January 28) shows what the agent was asked, what it used and where it went wrong. Endorsed mode, the default for new Threads since August, restricts answers to data the data team has marked as trusted. The agent became callable from a CLI and API on July 30, and evals arrived on August 4, so a data team can publish a suite of known questions and re-run it on a schedule to catch answer drift. Hex’s last disclosed round was a $70 million Series C in May 2025, led by Avra with a16z, Sequoia and Snowflake Ventures, bringing total funding to $172 million.

Can non-analyst ops people self-serve in Hex?

Yes, with one precondition: someone has to model the data first. A RevOps manager asking “how many accounts renewed below list price last quarter” gets a correct answer when renewals, list price and discount are defined in a semantic model and endorsed. Against raw CRM tables, the agent will write plausible SQL that picks the wrong amount field. That is the whole difference between a useful rollout and a credibility problem. Hex’s own credit guide prices the work: a simple count is about 1 credit, a basic chart about 2, a 12-month ARR forecast about 5, an interactive data app about 10.

Pricing reality

Community is free with 5 AI credits a month. Professional is $36 per editor per month with 30 credits per editor; Team is $75 per editor per month with 40 credits per editor, unlimited published apps, schedules and alerts. Explorer seats, the role for people who use Threads and apps without building, get 10 credits a month on Team; Hex does not publish their price. Enterprise is custom, with SSO via OIDC, audit logs and bring-your-own-key across OpenAI, Anthropic, AWS Bedrock and Google Gemini. Extra credits auto-top-up in 50-credit packs for $25, which is $0.50 a credit. Medium compute (8 GB, 1 CPU) is included; larger profiles bill per minute from $0.32 an hour.

Buyer data: Vendr puts the median contract at $38,550 a year across 84 purchases, ranging $12,726 to $122,700, with buyers landing about 41% below list. List price for five Team editors is $4,500 a year, so the Vendr median is mostly Enterprise deals carrying SSO and Explorer seats. Budget $5,000–$10,000 a year for a Team workspace and $25,000–$40,000 once you need Enterprise controls and dozens of Explorers, before warehouse compute.

Best for

A data or analytics lead at a company with a modeled Snowflake, BigQuery, Databricks or ClickHouse warehouse, who wants one place to do analyst-grade notebook work and also hand governed self-serve questions to RevOps and CS staff in Slack. Skip it if no one on staff writes SQL: Hex assumes a data team exists to model, endorse and audit.

Versus the alternatives

Power BI leads BI by installed base at $14 per user per month for Pro. Pick it when you are a Microsoft shop and the job is fixed dashboards. Tableau Cloud Creator lists at $75 per user per month; pick it when trained analysts do heavy visual exploration. Neither offers Hex’s notebook-plus-agent loop.

Sigma is the closest buying decision. Pick Sigma when spreadsheet-native business users must edit governed data, such as forecast submission or quota planning, because its writeback input tables have no Hex equivalent. Pick Hex when the work starts with an analyst’s question, needs Python, or should be answerable from Slack and Claude by people who never open a BI tool.

Omni is the fastest-growing entrant, with a $120 million Series C at a $1.5 billion valuation in April 2026. Pick Omni for a Looker-style governed semantic layer under classic dashboards. Pick Hex when exploratory analysis and conversational answers matter more than a dashboard catalog.

If you run on a single warehouse and only need text-to-SQL, test the warehouse’s own agent first: Databricks Genie and Snowflake’s Cortex Analyst run inside platforms you already pay for. Move to Hex when you need cross-source projects, notebooks and audit trails those tools do not give you.

Watch-outs

• Credits run out faster than seat counts suggest. A Team editor’s 40 credits cover about four data-app builds a month, and Slack, MCP and CLI runs all count against it. Guard: set a monthly spend limit on auto top-ups before you open Threads to Explorers, and review the agent activity breakdown in Context Studio after 30 days.
• Unmodeled data produces confident wrong answers. Guard: keep Endorsed mode on, model and endorse the 10–20 metrics RevOps actually asks about before launch, and publish an eval suite of those questions on a weekly schedule.
• Sensitive tables can leak through external clients. Guard: mark PII-bearing connections as sensitive so the MCP server and Slack agent cannot use them, and pair it with warehouse-level row security rather than relying on Hex alone.
• Explorer pricing is quote-only. Guard: get the Explorer seat price and its credit grant in writing before sizing a 50-person rollout, since Explorers will outnumber editors.