Skip to main content

​
Code Editor

On the left hand side of your screen, you have a code editor. This is where you define your report in markdown. The code editor is enriched for data development with:
  • Slash commands: Type a / to list components, accept suggestions with tab.
  • Autocompletion: For components and SQL fences, accept suggestions with tab.
  • Syntax highlighting: For markdown, SQL, and components.
  • Validation: Invalid code is underlined in red, hover to see errors, and click “Fix” for AI assistance.

​
Preview

As you modify the code in the editor, the preview will update in real time, including rerunning SQL queries and rerendering charts when you make changes. Nothing is published until you click the “Publish” button in the Share menu.

​
Evidence Agent

The right sidebar contains a chat interface with the Evidence Agent.
  • The agent is trained to generate valid Evidence markdown
  • It knows the schema of any data sources you have added
The agent can generate reports, fix errors, and answer general querstions about Evidence Studio.

​
Agent Capabilities

The agent has the following capabilities:
  • Generate reports: The agent can generate valid Evidence markdown, including the SQL queries and chart components to create reports from your data.
  • Edit reports: The agent can edit the markdown of your reports, including adding or removing components.
  • Read the schema: The agent can read the schema of your data sources, and use it to generate valid SQL queries.
  • Fix errors: The agent can fix errors in the editor, including invalid SQL queries and missing components.
  • Read documentation: The agent can answer general questions about Evidence Studio, including how to use the editor, and how to use the agent.
The Evidence Agent suggests changes in a “diff view”, which you can accept, reject, or send another message to improve.

​
AI Models

The Evidence Agent uses:
  • OpenAI GPT-4o: For general purpose reasoning.
  • Mistral Codestral Latest: For generating Evidence markdown.
Data from your data sources is never sent to these models. The only information sent to the models is the schema of your data sources, and relevant markdown of your reports, when you request it.