AI agents that work with your team’s definitions.
An agent gives AI Analyst your data and the context behind it: your definitions, SQL your team has reviewed, and what it learns from feedback. Use it wherever your team asks, and test its answers as your definitions change.
Free plan. No credit card required.
Give each team the context it needs
Choose the data an agent can use and the people who can use it. Give finance its revenue definitions and marketing its campaign conventions, each in its own agent.
- Data sources and tables
- Business definitions and reference material
- People with access
Start with Tune with AI, which drafts the agent from a few questions, or set it up yourself.
Finance agent
Month-end revenue, refunds and plan mix, in the finance team’s definitions.
A rule it reads: “For net revenue, subtract refunds issued in the period and exclude test accounts.”
Example agent configurations.
Marketing agent
Campaign performance and lead sources, using marketing’s attribution conventions.
A rule it reads: “Credit a lead to the first campaign it came from.”
Example agent configurations.
Support agent
Ticket volume, response times and recurring issues for the support team.
A rule it reads: “Measure first response time in business hours.”
Example agent configurations.
Define, reuse, remember, test
Four things shape how an agent answers. Each has its own place in the agent’s settings.
- Rule
Knowledge & rules
Explain your metrics, what your tables hold, and the steps for analyses you repeat.
Explore knowledge & rules - Example
Examples
Give familiar questions SQL your team has run and reviewed.
Explore examples - Memory
Memory
Review the context your agent keeps from conversations and feedback.
Explore memory - Eval
Evals
Test important questions and compare the results after every change.
Explore evals
Define, test, refine
Example: refining a Finance agent. Each change is a sentence, a query or a test.
Passing
September net revenue: $112,000
Rule clarified, example saved. Same evals rerun.
Sample data
Add a definition
A rule in plain language: “Revenue means net of refunds.” The agent reads it with every question.
About knowledge & rulesTest a question
Add “What was net revenue in September?” as an eval, with the SQL that answers it, and run it.
About evalsInspect an issue
The eval fails, and its note shows how the answer differed: it left refunds out.
About evalsRefine and rerun
Clarify the rule, save the reviewed SQL as an example, and run the same evals again to see what changed.
About examples
Use the same agent where you ask
Bring the agent’s definitions into AI Analyst and the places your team already works, so everyone starts from the same context.
| Where | How it uses the agent | Product |
|---|---|---|
| AI Analyst | Pick the agent for your question or analysis. Answers show the query they ran. | AI Analyst |
| Slack | Ask in a channel or a DM through the agent you choose. | AI Analyst |
| Claude | Choose the agent your Supaboard connection uses. | AI Analyst |
| Datasites | Questions on a datasite go through the agent you give it. | Datasites |
| Embedded dashboards | Viewers ask follow-up questions through the agent you set. | Dashboards |
| Automations | An Ask Analysts step answers through the agent you choose. | Automations |
Frequently asked questions
What is an agent in Supaboard?
The data and business context AI Analyst answers from. You choose its tables and people, give it definitions, reference material and examples, and refine it with evals and feedback.
Do I need to create an agent before asking a question?
No. Supaboard creates a default agent when your first data source is connected and its tables have been read, with notes on each table. Add more agents for different teams or subjects.
How do I check whether it answers correctly?
Add evals: questions with the SQL that answers them. Run them after you change the agent, and inspect any that fail.
Can different teams use different agents?
Yes. Each agent has its own data, context and people, configured for the team using it.
Can I see how an answer was worked out?
Every data-backed answer shows the query it ran. When that query matches one of your examples, the answer is marked as a verified query.
Does it work in Slack and Claude?
Yes. Slack and the Claude connection answer through the agent you choose, with its rules, examples and memory.
Build an agent around your team’s data
Start FreeBook a 30-minute demo
Free plan. No credit card required.