> For the complete documentation index, see [llms.txt](https://docs.rootcause.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.rootcause.ai/more-details/rootcause-assistant.md). # RootCause Assistant (BETA) The RootCause Assistant lets you interact with your data, Digital Twins, and analyses through natural language. Ask questions, run simulations, and explore causal relationships — all through conversation. Unlike typical AI chat tools, the Assistant doesn't try to reason about your data directly. It orchestrates RootCause's analytical engines — causal discovery, counterfactual simulation, data queries — and reports what those engines compute. The AI handles the conversation; the math comes from purpose-built algorithms.

Four-stage diagram showing how RootCause Assistant works: You Ask, Assistant Orchestrates, Engines Compute, Result plus Evidence — The Assistant is an orchestration layer. Your question triggers precise tool calls to causal engines — the numbers come from the math, not from AI pattern-matching.

*** ### Accessing the Assistant Click the chat button in the platform toolbar, or press `⌘⇧J`. The Assistant panel docks on the right side of the screen. It is available inside a workspace — on Sources, Datasets, Ontology, Digital Twins, Applications and Reports — but not on organisation-level pages, and not on the step-by-step wizards (twin build, import and export, new application, ontology binding), where the toolbar has no chat button. It adapts to your current context: on a Digital Twin page it can run simulations directly; on a Report page it can help add content. The panel opens with suggested prompts drawn from the workspace, and its controls include **New chat** to start over and **Full screen** to expand the conversation. Drag the panel's left edge to resize it, anywhere from 260 pixels wide up to 320 pixels short of the window's width, so the page you are working on always keeps a column of its own; the width is remembered between visits. Dragging the edge a further 100 pixels past the minimum closes the panel rather than stopping there, and opening it again restores the width it had before the drag. *** ### What you can do **Ask questions about your data** > "What's the average churn rate across all customers?" "Show me the distribution of customer tenure" "What columns are in the Sales dataset?" The Assistant executes queries against your data and returns results as tables, charts, or summary statistics — each with an evidence reference you can verify. **Run causal simulations** > "What happens to churn if we offer free tech support?" "Find the optimal marketing spend to maximize conversions" "What's driving customer churn?" "What would it take to reduce churn by 20%?" The Assistant calls the appropriate simulation type — prediction, intervention, optimization, best action, explanation, root cause analysis, anomaly scan, or forecast — runs it against your Digital Twin, and presents computed results with full traceability. **Navigate the platform** > "Open the Revenue Digital Twin" "Show me the customer 360 Dataset" "Take me to last month's board report" **Get help** > "How do I upload a file?" "What's the difference between intervention and counterfactual?" "Explain what the causal graph is showing me"

RootCause Assistant conversation answering What's the average churn rate — 26.3%, from 1,313 churned out of 4,998 total customers — with a query results table attached as evidence — A data query in the Assistant: the average churn rate is 26.3% — 1,313 of 4,998 customers churned — with the query results table attached as clickable evidence.

*** ### Evidence and traceability Every result from the Assistant carries an evidence reference. Click it to see the query or computation that produced the result, the source data used, the Digital Twin version, and the simulation parameters. **Example:** > **You:** What's the predicted impact of a 10% price increase on revenue? > > **Assistant:** Based on the Revenue Twin, a 10% price increase is predicted to reduce revenue by 4.2% (95% CI: 3.1%–5.3%). The cited number itself is the evidence link: it renders underlined and opens the evidence behind it when clicked. For important decisions, click through to the evidence. Verify the query was correct, the data was appropriate, and the computation makes sense. *** ### Context mentions Use `@` to reference specific objects in your workspace: * `@Customer_360` — a Dataset * `@Revenue_Twin` — a Digital Twin * `@Q4_Report` — a report > "Run an intervention on @Revenue\_Twin to see what happens if we increase marketing spend by 20%" Context mentions eliminate ambiguity about which object to use for a query or simulation. *** ### Rich responses The Assistant returns structured results, not just text: data previews, charts, KPI cards, causal graphs, simulation results, and tables. Each includes an evidence link back to the underlying data or computation. *** ### Session context Within a session, the Assistant remembers what you have discussed — follow-up questions can reference earlier results without repeating yourself. Conversation history is also preserved across sessions: the Assistant can search your past conversations in the workspace and retrieve their details, so you can pick up where an earlier session left off. *** ### Tips * **Be specific** — "What's the average revenue per customer in Q4 from @Sales\_Data?" gets a better result than "Tell me about revenue" * **Use `@` mentions** — reference specific objects to avoid ambiguity * **Ask one thing at a time** — break complex requests into steps * **Check the evidence** — for decisions that matter, click through and verify *** > **BETA:** The RootCause Assistant is in active development. Capabilities are expanding and some responses may require verification.