Changelog

⚡ Fast mode​

Need an answer now? Flip on fast mode in the model picker to push Opus models to 2.5x faster token output, at increased usage.

The boost is most dramatic on long blocks of pure token output: big explanations, large blocks of code, or heavy reasoning. Questions that require a lot of querying and tool calls will still feel faster, just less dramatically, since a chunk of that time is calls and execution rather than token generation.

🧠 GPT-5.6 Sol, Terra, and Luna​

OpenAI's GPT-5.6 family is now in the model picker:

• Sol for complex, multi-step analytical work
• Terra for balanced everyday use
• Luna for fast, low-cost lookups

Select one when you're prompting the agent, or keep Auto on, which automatically selects the model best suited for the task.

✅ Default new Threads to Endorsed mode​

New threads from the Hex UI or Slack can now open in Endorsed Mode by default, so Editors land in endorsed projects and data without anyone having to remember to flip a switch. Admins can toggle off this enforcement in Settings > AI & agents.

🏷️ See which model Auto picked​

When the agent runs on Auto, Context Studio now shows which model it picked, shown as Auto (Opus 4.7), making it easy to see what Auto chose when reviewing usage and cost.

🧪 Introducing Evals​

Measure Hex Agent performance and test context changes safely before they ship, all from the CLI.

If you've added a guide or tweaked a semantic model, Evals helps you measure whether it actually helped — before you publish. To do that, define a suite of test questions with expected outcomes and run them against the Hex Agent.

Evals turn agent quality and cost into something that you can measure and improve, so you're not just going on vibes.

Some ways teams are using it:

• Test context changes before publishing by forking context and running Evals to validate agent improvements.
• Catch drift before your users do by running the same Evals against production on a cadence.
• Run configuration sweeps to compare cost and pass rate side by side.

Learn more about Evals or dive into the docs for the full rubric reference.

Getting started: Install the Hex CLI and follow the bootstrap prompt for Claude Code, Cursor, or your coding agent of choice — it scans your recent Threads, clusters them into your team's most common question topics, and drafts a starter suite from real usage instead of a blank file.

🤖 Hex Agent via CLI and API​

You can now have conversations with the Hex Agent via CLI and API, so you can set it to work from the tools and scripts you already use, not just the browser. A few things that unlocks:

• Kick off a thread with the agent right from the command line by running hex thread create — the agent finds the tables, builds the analyses, and hands you back a link to the project without any separate coding tools.
• Wire the agent into where your teams already working via API — you can ask questions in your coding agent, in your terminal, or whatever other app you’re working in, and get detailed analysis powered by the Hex Agent.
• Migrate off your old BI tool by pointing the agent at a dashboard and having it rebuild the thing as a Hex project instead of porting queries by hand.

Ready to get started? If you have enabled API access enabled, just paste the following in your terminal (or check out the documentation here)

curl installer: curl -fsSL https://hex.tech/install.sh | bash

homebrew installer: brew install hex-inc/hex-cli/hex

💻 Cell outputs in the CLI​

If you've ever run a cell and wondered "wait, what did that output actually look like?" This is for you.

The Hex CLI can now show you what a cell actually returned, right in your terminal, so you don't have to run something and then flip back to the browser to see what happened.

Add --with-output to your next hex cell run or hex cell get.

Fetching a big table that won't fit a terminal window? The CLI will switch between a horizontal and vertical layout depending on the space it detects when returning results. We also added the vertical layout to all CLI commands that render a table, with new --expand and --no-expand flags.

🔗 See which resources the agent used​

The Hex Agent can pull from a ton of context — guides, tables, semantic models — and until now it was hard to see what it actually reached for. Now Hex shows you the resources the agent used, and even distinguishes whether it just looked at a resource or actually used it in the analysis.

It shows up in three places:

• Open the Thread Inspector in any thread to see its resources.
• Head to the Context Studio to see which of your guides and tables are actually earning their keep across your workspace, so you know what to invest in and what to prune.
• Use an agent to analyze the data via CLI and **the resources in the output.

So instead of guessing whether the agent found the right guide or leaned on the wrong table, you can see it, and tune your context accordingly.

🧠 Claude Opus 5​

Claude Opus 5, Anthropic's newest and most capable model, is now in the model picker. It's the strongest option in Hex for complex, open-ended analytical work, and it holds up impressively well even at lower effort levels.

It's a premium model, so use it for your hardest questions rather than routine lookups. Pick it in personal settings, or switch to it when you prompt the agent.

🔄 Faster schema refreshes for Snowflake​

For Snowflake, a refresh now fetches only what's changed instead of all objects in your warehouse. For a typical Snowflake connection, this means we’re refreshing only 30% of the tables in the schema browser. Some customers have experienced up to 10x faster schema refresh times as a result of this change.

With this change, and other improvements we’ve made to our system behind the scenes, we’ve made refreshes significantly lighter on our system. Every data connection, regardless of warehouse, benefits from these changes — if your refresh has ever been stuck “queueing”, you’ll see this happen far less often.

🛠️ Other improvements​

• Quicker simple charts from the agent: The Hex Agent now builds straightforward charts directly instead of handing off to a separate step, so simple visualizations show up faster. It still brings in the specialist for complex, highly customized charts.
• Search settings: Hit Cmd + K from settings to jump straight to the setting you’re looking for.
• Configure how the Hex Slack integration treats sensitive connections: Previously, we had a setting that excluded a data connection from use for the Slackbot and MCP. Now, Admins can choose how the Slack integration treats those connections: the connections can be ignored completely (same as today), or they can be used, but the Agent will only send a link to the thread, rather than posting directly to Slack.

✏️ Edit generative app code in Hex​

Sometimes you don't need the agent, you just need to change one line. Previously, you had to prompt the agent every time you wanted to make a simple fix. Now, generative apps can open in a real code editor right inside Hex, so when you want to change a label, adjust some styling, or fix a bit of logic, you can just do it.

Your edits sync live and save as you go. The agent still manages the app manifest behind the scenes, so the file stays read-only.

Open any generative app, switch into Code view, and pick a file from the sidebar to start editing.

⬇️ CSV downloads for every chart and table​

Charts and tables in generative apps can now export to CSV, so the people looking at your app can pull the numbers behind it without coming to you for the raw data.

The download gives viewers exactly what they're looking at, with current filters, sorting, and search carried through. It respects your workspace's download setting, so if an admin has downloads off the button never appears, and every download is audit-logged. Dates come out as clean ISO timestamps, so they land correctly in whatever tool someone opens next.

New charts and tables get it automatically. But for an app you’ve already built, just tell the agent: "Add Download CSV to the charts and tables in this app."

🔔 Desktop notifications for the agent​

The Hex Agent can now send a desktop notification when it finishes a run or needs a decision from you, so you can fire off a prompt, go work on something else, and get pulled back the moment the agent needs you.

It only pings when you've switched away from the thread, so if you're watching it, nothing pops up. Kick off your next agent run and accept the prompt when it appears, or turn it on anytime in personal settings under Notifications.

📸 The agent can see the apps it builds​

The Hex Agent can now take a full-page screenshot of a generative app as it builds, and zoom in on a single chart or view, so it can check its own work and fix what looks off before handing it back to you.

You won't touch this one directly, you'll just notice apps come out looking right more often.

⤴️ Spend limits for auto top-ups​

Forecasting credit usage is notoriously difficult, but staying in budget should be a breeze.

Hex supports workspace credit pool auto top-ups with an optional monthly spend limit. This means you only pay for what you need, and you never go over budget.

👥 Credit usage controls​

Auto top-up spend limits are great, but what if you purchased credits upfront for the whole year? And how do you prevent one power user from running away with the whole budget?

Admins can now set per-user-per-month add-on credit limits (unlimited credits, no access, or a custom limit) applied to individuals, groups, or workspace default.

An individual limit always wins the argument, then the highest group limit that applies, then the workspace default if nothing else was set.

For the full picture on how credits and usage visibility work in Hex, read the blog. And if you want tactical advice on managing spend, our Head of Data Katie wrote up how her team approaches credit efficiency.

🧵 Access Context Studio data via CLI​

Data on credit consumption, topics, and warnings is already available in Context Studio, but we know you want to slice and dice this as you please.

Now you can pull it via CLI or API, beyond just exploring through the UI.

The export is easy, the useful part is what you do next:

• Analyze who's doing what across your workspace
• Identify opportunities to strengthen your context
• Understand which topics users spend credits on

Here's what's available to pull: every thread in your workspace, the full back-and-forth of user and agent messages, and all the thread-level metadata that comes with it. Join it with user details like name, email, and group membership to finally put a face to a thread.

Use what you learn from agent conversations to review and refine the context that shapes future responses.

🔓 Enterprise user role request controls​

Enterprise admins can stop fielding "hey, can you bump me to Editor?" messages and now control how users request workspace role upgrades in Hex. Maybe it's a Viewer who wants to build their own explorations, or an Explorer who's hit a wall without CLI access and wants to start automating their own work.

Head to Settings > Users > Role requests to add custom guidance that appears when someone requests a role upgrade, or turn off role requests entirely with a custom message.