Three surfaces. One workspace.
Jubi Studio is the surface people use. The Assistant is for open-ended questions over your data. The dashboard widget sits inside Metabase and already knows what you're looking at. Context Studio is where analysts curate the semantic layer that keeps the AI honest.
Three reading patterns. One product.
A business user, a Metabase user, and an analyst each see Studio in a different form. Every surface is shaped to how that person actually works, not how an interface designer wishes they did.
Conversational. Threads persist across sessions. Web search and file uploads are available when the question reaches beyond your warehouse. Built for follow-up, not one-shot prompts.
- Persistent threads
- Web search on demand
- File & image attachments
- Citations to source cards
- Replayable per request
- Permissioned to the user
Open the Jubi tab on any Metabase dashboard. It already knows which dashboard you're on, has pre-loaded the data behind each card, and is ready for the next question without context-switching.
- Dashboard-aware
- Card-level grounding
- Drill-down inline
- No context switch
- Inherits Metabase auth
- Shareable as a follow-up
The annotation surface for the people who maintain the data. Annotate the data model, scope workspaces with their own playbooks, glossary terms, and verified metrics. The AI gets sharper as analysts contribute.
- Semantic catalog
- Workspace scoping
- Metric definitions
- Glossary terms
- Playbooks
- Coverage tracking
Workspace semantic-layer editing (glossary terms, metric definitions, playbook content) is in active build. Today, Context Studio supports browsing the data model and scoping workspaces; the editing UI for the semantic layer is being added in the next release.
Four people. Same workspace.
Studio is shaped to four jobs that already exist in every enterprise. It doesn't ask people to change how they work — it fits the way they already do.
Anyone who needs to know something about the business and can't or won't write SQL. They open Studio, ask in plain language, and get an answer with a citation.
- Ad-hoc questions in the Assistant
- The widget alongside the dashboard they already trust
- No SQL, no dashboard-build queue
The people who already maintain the dashboards and know what the columns actually mean. They use Context Studio to teach the platform that knowledge once instead of answering the same question forty times.
- Curate the semantic catalog
- Define verified metrics and glossary terms
- Validate AI-generated queries before they go wide
Teams running repeatable analyses — weekly reviews, exception monitoring, ops dashboards. Studio's playbooks let them codify a question once and run it on a cadence.
- Playbooks for repeated analyses
- Multi-step agent runs (roadmap)
- Hand-off to alerting and routing
Security, compliance, and platform owners. Studio surfaces what's happening; Guardian decides what's allowed; Atlas defines what the answers mean. Admins live across all three.
- Audit and replay every request
- Configure SSO group → workspace permissions
- Track coverage and adoption
What Studio reads. What Studio writes.
The boundary is deliberate. Read paths are wide so that questions land. Write paths are narrow and explicit so that nothing is changed by accident.
- Metabase dashboards, saved questions, and collections you have permission for
- Query results from any database connected to Metabase, under your access rights
- Atlas semantic layer entries scoped to the workspace
- Conversation history within the same session
- Files you upload during the conversation
- Create or update Metabase dashboards and cards on your behalf
- Add to the conversation thread (always)
- Save a playbook or workspace artefact (when permitted)
- Workspace semantic-layer edits (planned, gated by analyst role)
Studio reaches. Guardian protects. Atlas grounds.
Every question typed into Studio takes the same path. The user sees one thread; the platform sees a request, a policy decision, and a grounded answer.
Studio is the reach — the surface where the user is. Guardian is the gate — every request, every answer, every audit. Atlas is the ground — the layer the AI reads from instead of guessing. None of the three works alone. See the platform architecture for the full diagram.