MinusX Agent
A trainable, trustworthy agent across the entire product surface
The MinusX agent is not a chatbot bolted onto a BI tool. It's a first-class part of the platform — present on every page, aware of your current context, and capable of taking real actions. Every action produces a visible, editable artifact. Nothing is hidden.
What it can do
| Page | What can you do? | Examples |
|---|---|---|
| Explore | Go from question to insight in seconds | "What are the top 10 customers by revenue?" "Now filter to Q1" "Show it as a bar chart" |
| Questions | Build production-grade analyses without writing SQL from scratch | "Add a date filter" "Change the aggregation to AVG" "Make this a pivot table with region as rows" |
| Dashboards | Assemble a live view your team actually checks every morning | "Add a revenue chart" "Create a filter for region" "Move the KPI cards to the top" |
| Reports | Make the data come to people instead of the other way around | "Change the schedule to weekly" "Add the marketing team" |
| Alerts | Know when something breaks before your customers do | "Alert me when revenue drops below $10k" "Change to daily checks" |
| Slack | Let your whole team ask data questions where they already work | "@minusx what was revenue last week?" "@minusx top 5 churned accounts this month" |
| MCP | Give your AI coding tools direct access to your data | Search schemas, run queries, list connections — from Claude Desktop, Cursor, or Windsurf |
Context awareness
The agent reads three layers of context before every action:
- Knowledge Base — whitelisted tables, metric definitions, business rules
- Current page — the SQL, chart, or dashboard you're looking at right now
- Conversation history — what you've discussed in this session
This is why it understands "now break that down by region" — it knows what "that" refers to.
Trainability
The agent gets better as you invest in context:
- Add table context → the agent uses the right tables
- Write text context → the agent understands your metrics and conventions
- Run evals → you measure accuracy and find gaps
- Iterate → improve context based on eval results
This is the core loop. The agent is only as good as the context you give it.
The agent reads your Knowledge Base before every query. Maintaining good context is the single most impactful thing you can do for accuracy.