Enable Scout AI for your workspace and give it the context, data access and models it needs
Scout works with its default settings. When it is enabled, it starts to analyze the monitors, tests and issues in Coalesce Quality. You can change how Scout behaves with triage settings, business context and the data sources it can query.
Go to Settings → Scout AI. If Scout is not enabled, an Admin clicks Enable Scout AI on this page. If you do not see the page, contact the Coalesce Quality team to turn Scout on for your workspace. The setup asks for the deployment method and the data sources, see Infrastructure.The settings page has these sections: Context, Infrastructure, Triage, Testing, Models and Danger zone.
Tell Scout about your business and data, the same as you would for a new analytics engineer. In About your business, describe your critical datasets, workflows and known quirks in your data model.Example: “We operate a B2B payment provider solution. Our key datasets are around payments and customer orders, especially if they’re used for billing. We monitor many variables across key metrics, which should be assumed independent unless proven otherwise.”
Scout uses metadata, and direct access to your data sources, to run diagnostic and profiling queries.
Deployment method: select Coalesce Quality managed (recommended) or Self-hosted (on-premises). In managed mode, Coalesce Quality makes the LLM calls and sends the data warehouse queries for Scout. In self-hosted mode, you run Scout, the LLM endpoint and the warehouse connections yourself. When you select self-hosted, this section shows the client ID and client secret for your deployment. See On-Premise Setup.
Data sources: select the data warehouse integrations that Scout can query for profiling, sampling and diagnostics. If you leave out an integration, Scout cannot read its data, and its suggestions for those assets are less accurate.
Auto triage settings: choose which assets and severity levels Scout triages automatically. For example, triage Error and Fatal issues, but leave out Warnings to reduce noise.
Triage instructions: tell Scout how to triage issues, group them into incidents and prioritize them by business impact.For example: “Group issues only when there is clear evidence of correlation.” or “Issues on the payments schemas are critical as they may signal customer-facing problems.”
Issue grouping: tell Scout which criteria to use to group issues, for example the severity or the type of the issue.
Scout can suggest changes to the prompts in Context, Triage and Testing. When a suggestion is ready, the section shows a banner. Open it to compare the current prompt with the suggestion. You can edit the suggestion before you accept it, or dismiss it. A dismissed suggestion does not change your settings.
Under Settings → Scout AI → Models you can change the models that Scout uses: a thinking model for reasoning and a summarization model for summaries. The options include stable and latest model versions. Opus-class models are available on request.Each model in the list shows where it runs for your workspace: your region (for example EU), a location outside your region (Singapore, for Claude models in the AU region), or Global for a model served from the Google Vertex AI global endpoint. The location of the selected model also shows under the picker. See AI Processing for where each region runs inference.The global endpoint can process requests on any continent, so models served from it are off by default. If your region offers such a model, an Admin can turn on Allow models served from the global endpoint in the same section. When the setting is off and the selected model needs the global endpoint, Scout uses the default model.
