Context Studio
- Available on the Team and Enterprise plans.
- Users need the Admin or Manager role to access Context Studio.
- Only Admins can view individual conversations.
Context Studio enables Admins and Managers to observe, evaluate, and improve how Hex agents behave across your workspace. It combines observability and context management in one place, giving data teams the tools they need to confidently provide a solution where anyone in their organization can answer data questions using Hex.
As agent usage grows, Context Studio gives you visibility into where agents perform well and where they struggle, then provides direct access to the context settings that influence their behavior. For guidance on configuring these context sources, see setting up your workspace for AI agents.
Dashboard: Observe agent behavior
The Dashboard shows how agents are being used across your workspace. You can track conversation volume, identify your most active users, and spot patterns in feedback to understand where agents excel and where context improvements could help.
You can filter by agent type (Threads, Notebook agent, Semantic Workbench), workspace role, individual users, source location (inside Hex, Slack or MCP) and time range (7, 30, or 90 days).
Admins can open individual agent conversations to review the agent's reasoning, see what context it referenced, and identify where improvements are needed. These detailed conversation views are diagnostic tools available only to Admins. Managers see aggregated metrics but cannot view individual conversations.
Context sources: Experiment with context updates
The Context sources tab provides quick access to the settings that influence agent behavior. Context Studio brings them together so you can review and adjust them based on what you observe in the Dashboard.
Agent workspace rules file
Define workspace-wide guidance for agents, including high-level business terminology, data conventions, and preferred practices. Edit your rules file directly in Context Studio. See Agent workspace rules file.
Semantic projects
Semantic projects provide the shared definitions agents use to interpret metrics, dimensions, and relationships in your data. Well-defined semantic models are one of the most effective ways to improve agent accuracy. Review which semantic projects are active in your workspace. See Introduction to semantic models.
Endorsements
Endorsements signal which data sources are most trusted in your organization. Agents prioritize endorsed tables and data objects when answering questions, reducing the chance they'll reference outdated or incorrect data. Review and update endorsements based on agent behavior. See Endorsed statuses.
Curating warehouse context
Beyond the Context Studio, agent accuracy also depends on well-curated warehouse metadata - like table and column descriptions, and schema filtering to exclude irrelevant data. For guidance on best practices, see Warehouse curation.