13. Chatbot And Assistant UI

Learning Goal

Build assistant-style interfaces that work alongside dashboards.

Why Add A Chat UI

An assistant UI is useful when users want to ask:

• why is this pipeline delayed?
• which dataset is stale?
• what changed since yesterday?
• what should I investigate first?
• summarize the current incidents

BrickflowUI gives you the building blocks. You decide whether the response comes from simple rules, an API route, a model, or a Databricks job.

Basic Chat

prompt, set_prompt = db.use_state("")
answer, set_answer = db.use_state("Ask me about pipeline health.")

db.Column(
    [
        db.ChatMessage("assistant", answer, name="Ops Copilot"),
        db.ChatInput(
            value=prompt,
            on_change=set_prompt,
            on_submit=lambda text: set_answer(f"Investigating: {text}"),
        ),
    ],
    gap=4,
)

Chat With Dashboard Context

rows = filtered_rows

def answer_question(text: str):
    if "failed" in text.lower():
        failed = [row for row in rows if row["status"] == "failed"]
        set_answer(f"{len(failed)} pipelines failed in the current filter.")
    else:
        set_answer(f"{len(rows)} pipelines match the current dashboard filters.")

Then:

db.ChatInput(value=prompt, on_change=set_prompt, on_submit=answer_question)

Assistant Layout Pattern

db.Card(
    [
        db.SectionHeader("Pipeline Copilot", "Ask questions about the filtered dashboard state."),
        db.ChatMessage("assistant", answer, name="Ops Copilot"),
        db.ChatInput(value=prompt, on_change=set_prompt, on_submit=answer_question),
    ],
    elevated=True,
)

Common Mistakes

• Building a chat box that ignores current dashboard filters.
• Not storing prompt text in state.
• Not clearing or updating the prompt after submit when desired.
• Making the assistant sound authoritative when the data is incomplete.

Exercise

Create a chat card that answers:

• "what is at risk?"
• "what costs most?"
• "how many pipelines are healthy?"

Use simple Python logic over mock rows.

Checkpoint

You should be able to create a useful assistant UI that is grounded in dashboard state.