AI Assistant

The assistant is a chat built into the dashboard. Ask it a question about your game and it looks up the answer in your project: it finds the right events, builds and runs the query, and explains the result. It also makes recommendations, like which events to add, what to test next, or which values are risky to change.

It can also make changes: edit configurations, start or stop experiments, and save charts to dashboards. Nothing changes until you approve it. Each proposed change appears in the chat as a card showing what will change, and it runs only when you click Approve. See Changes need your approval.

Each message costs a small amount of AI usage, paid from your organization's credits. See AI Credits & Budgets.

Open the assistantCopy Anchor Link

There are four ways in. All of them work on the project and environment you currently have selected.

• The Assistant page. Pick Assistant in the project sidebar for a full-page chat.

• The command bar. Press Cmd+K (Mac) or Ctrl+K (Windows), then press Tab or pick Ask AI assistant. Anything you already typed is sent as your first message. Press Esc, or Tab or Backspace in an empty prompt, to go back to search. Use the expand button to move the chat to the full Assistant page.

• Ask AI buttons. Dashboards, saved queries, configurations, experiments, deeplinks, and Community Pulse have an Ask AI button in the page header. It opens the assistant with that page already attached as context, and offers suggested questions. Some pages also show a row of suggestions under the header, which you can dismiss.

• Links. Any dashboard URL with a ?prompt= parameter opens the assistant with that text in the prompt, ready for you to review and send. It's never sent automatically. This is useful for sharing a question with a teammate or linking from your own tools:

  Copy Text

  https://dashboard.gamebeast.gg/live/organization/<orgId>/application/<projectId>/assistant?prompt=Which%20levels%20have%20the%20biggest%20drop-off%3F

  An organization-level link (…/organization/<orgId>?prompt=…) opens the organization's first project.

Press Enter to send and Shift+Enter for a new line. While the assistant is replying, the send button becomes a stop button.

What it can doCopy Anchor Link

When you don't give a time range, the assistant uses the last 30 days. Ask for "all time" if you want everything.

Times like "today" and "this week" use your project's timezone.

Results you can keepCopy Anchor Link

Charts and tables appear inline as cards. From a card's header you can:

• Open in query builder to keep editing the query in Explore.
• Add to dashboard to save it to one of your dashboards.
• Export it as CSV or PNG.

A cohort the assistant builds appears with its definition and member count, and Save Cohort and Edit Cohort buttons. Saving it is your action, using your own permissions. The assistant doesn't save it for you.

Experiment results show each group against the control, with the lift, a confidence interval, and whether the difference is significant. The assistant also warns when groups are unevenly sized, which can mean the results are unreliable.

Changes need your approvalCopy Anchor Link

When you ask for a change, the assistant reads the current state first and then proposes it. The chat pauses on a card showing:

• The reason the assistant gave for the change.
• What will change. For a configuration edit, that's a diff against the live configuration. For an experiment, it's the groups, the split, and each group's changes.
• Whether it goes live. Changes to your project's primary configuration are pushed to game servers as soon as you approve. The card marks these, in red in production.

Click Approve to apply the change, or Deny and optionally say why. Either way, the assistant carries on from your answer, and it's instructed not to propose a change again once you've denied it.

The assistant can:

• Set or remove values in a configuration (up to 25 per change).
• Create a configuration, delete one, or make one the primary configuration.
• Create an experiment, end an experiment, or put a specific player or server into an experiment group for testing.
• Save a chart it built as a query (optionally on a dashboard), change or delete a saved query, and create, rename, reorder, copy, or delete dashboards. Removing a query from a dashboard keeps the query; deleting a dashboard keeps its queries.

Some checks keep changes safe:

• Approve is disabled when the change can't be applied as shown. For example, someone else edited the configuration after the assistant read it, or you don't have permission to make the change. Ask the assistant to try again and it will re-read the configuration.
• Keys controlled by a running experiment are locked. A change to one is refused, even after you approve it, until the experiment ends.
• Private keys stay private. The assistant can change their values, but only you can mark a key private or public, in the editor.
• The primary configuration can't be deleted, and neither can a configuration a scheduled or running experiment is based on.
• Ending an experiment is immediate and permanent. Every player goes back to the base configuration, and the experiment can't be restarted. The card warns you before you approve.

Choose a modelCopy Anchor Link

The model picker next to the send button offers three models:

Beast and Godfrey can also think before they answer. The Thinking slider in the same picker goes from Off to High (the default is Medium). More thinking gives better answers to hard questions, but replies take longer and cost more. For simple lookups, turn it down or use Mini Beast.

Your choice of model and thinking level is saved in your browser and applies to every message you send until you change it.

Attach contextCopy Anchor Link

Use the + button in the prompt to pin things from your project to the conversation, so the assistant starts from exactly what you mean instead of searching for it. You can pin:

• Saved queries and dashboards
• Configurations: a whole configuration, a folder inside it, or one field
• Markers, or one of a marker's arguments
• Experiments, or a single experiment group
• Cohorts and deeplinks
• The Community Pulse overview, or one topic

Pinned items appear as chips above the prompt and stay attached for the rest of the conversation until you remove them with ×. You can pin up to 8 items per message. The Ask AI buttons around the dashboard pin the page you're on for you.

If you pin something you don't have permission to see, or that has since been deleted, the assistant is told it's unavailable.

ImagesCopy Anchor Link

The same + menu can attach files. The assistant accepts images (PNG, JPEG, WebP, and GIF), such as a screenshot of a chart or a mockup of your UI. You can attach up to 5 files per message, up to 15 MB each and 30 MB in total.

ConversationsCopy Anchor Link

Conversations aren't saved. A chat lasts until you reload the page or switch organization, project, or environment. Switching starts a fresh chat so the assistant never mixes data from two projects. If there's an answer you want to keep, save the chart to a dashboard, export it, or copy the text.

PermissionsCopy Anchor Link

The assistant can do only what you can already do in the project. It acts with your role and project access, which are checked on every tool it uses:

• Anyone who can view a project can use its assistant.
• The assistant is only offered tools your role allows. If your role can't read experiments, it can't read them in chat either.
• To approve a change, you need the same permission as for the same action in the dashboard. For example, editing configurations, or managing project settings to change the primary configuration.

Analytics answers also need your organization's billing to be in good standing, just as dashboards do.

CostCopy Anchor Link

Every message is charged for the AI usage it takes, from your organization's credits. Longer conversations, bigger models, and more thinking cost more. You can see what the assistant has cost under Billing → Overview, and set a monthly cap on AI spend there. The full details are on AI Credits & Budgets.

TroubleshootingCopy Anchor Link

The assistant says your balance is empty. Your organization has run out of credits. Pick Reload Balance to add more, or Go to Billing. See When you run out.

The assistant says your AI budget is exceeded. Your organization's monthly AI cap has been reached. Someone with billing access can raise or remove it from Manage AI Budget. The assistant becomes available again straight away. See Set an AI cap.

Godfrey is locked. Godfrey requires a paid plan. Mini Beast and Beast are available on every plan.

The assistant is temporarily unavailable. We've paused the assistant, usually during an incident with an AI provider. Your data is unaffected. Try again later.

A message failed. Use Retry under the error. If it fails again, start a new message with a narrower question, or use a different model.

The assistant can't find something I know exists. Check that you're in the right project and environment. The assistant only sees the one you have selected. Pinning the item with + removes the guesswork.

An answer looks wrong. Open the chart in the query builder to check exactly what was run. You can also expand the Worked for… row above the answer to see each step the assistant took.