> ## Documentation Index
> Fetch the complete documentation index at: https://docs.buildbetter.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Agent Chat (BETA)

> AI-powered research agent that investigates your data, plans analysis, and delivers evidence-backed insights

Agent Chat is BuildBetter's next-generation AI assistant. Instead of simple Q\&A, it acts as a research agent—planning multi-step investigations, searching across your entire workspace, and delivering comprehensive analysis with every claim backed by evidence you can verify.

Agent Chat can also help manage tracked items and projects. It can create, update, merge, promote, and organize related evidence and links when your workspace has the required permissions. Destructive project-management actions require explicit confirmation before they run.

## What Makes Agent Chat Different

<CardGroup cols={2}>
  <Card title="Research, Not Just Answers" icon="flask">
    The agent plans its approach, searches multiple data sources, cross-references findings, and builds a complete picture before responding
  </Card>

  <Card title="Evidence-Backed" icon="link">
    Every insight includes clickable citations to the original call, signal, or document so you can verify and dig deeper
  </Card>

  <Card title="Deep Analysis" icon="microscope">
    Run multi-phase investigations that analyze patterns across hundreds of conversations, not just surface-level keyword matching
  </Card>

  <Card title="Context-Scoped" icon="crosshairs">
    Ask questions scoped to a specific person, company, document, folder, or signal for precise, focused results
  </Card>
</CardGroup>

## How It Works

When you ask Agent Chat a question, it doesn't just search and summarize. It:

1. **Plans the investigation** — Breaks your question into research steps and determines which data sources to query
2. **Searches your workspace** — Queries calls, signals, documents, people, and companies across your entire dataset
3. **Cross-references findings** — Connects patterns across different sources, time periods, and customer segments
4. **Synthesizes results** — Delivers a structured analysis with evidence, not just a list of matches
5. **Cites everything** — Every claim links back to the original source so you can click through and verify

<Tip>
  You can watch the agent work in real-time. As it researches, you'll see which tools it's using and what data it's pulling—full transparency into how it reaches its conclusions.
</Tip>

## What You Can Do

### Deep Customer Research

Go beyond simple searches to understand what's really happening with your customers:

* "What are the top 5 reasons enterprise customers churned in Q4, and what did they say in their last calls?"
* "Compare how our biggest accounts talk about pricing versus how mid-market customers do"
* "What patterns do you see across all calls with \[Company Name] over the past 6 months?"

### Cross-Source Analysis

The agent connects insights across your entire workspace:

* "What feature requests keep coming up in both support tickets and sales calls?"
* "Find every time a customer mentioned \[Competitor] and analyze the context"
* "What commitments did our team make to customers this month that haven't been tracked?"

### Person & Company Intelligence

Scope your questions to specific people or organizations:

* "What has \[Person Name] talked about across all their calls?"
* "Give me a complete picture of \[Company Name]'s feedback history"
* "What are the common themes from calls with our healthcare customers?"

### Strategic Analysis

Ask complex questions that require connecting multiple data points:

* "What's the sentiment trend across our top 20 accounts over the past quarter?"
* "Which product areas generate the most negative feedback and what are the root causes?"
* "Based on recent calls, what should our product team prioritize next quarter?"

## Scoping Your Questions

Agent Chat can focus its research on specific contexts for more precise answers:

| Scope        | How to Use                   | Example                                               |
| ------------ | ---------------------------- | ----------------------------------------------------- |
| **Person**   | Ask about a specific contact | "What has Sarah Chen discussed across all her calls?" |
| **Company**  | Focus on one organization    | "Summarize all feedback from Acme Corp"               |
| **Document** | Analyze a specific document  | Drag a document in and ask questions about it         |
| **Folder**   | Research a collection        | Drag a folder in to analyze everything inside it      |
| **Signal**   | Dig into a specific insight  | Click into a signal and ask for deeper context        |
| **Call**     | Deep-dive a recording        | Drag a call in for detailed transcript analysis       |

<Note>
  You can also drag multiple items into the chat to analyze them together—compare calls, cross-reference documents, or find patterns across a collection.
</Note>

## Tools the Agent Uses

The agent has access to a suite of research tools that it selects automatically based on your question:

* **Workspace Search** — Full-text and semantic search across all your data
* **Signal Analysis** — Query and filter your extracted signals by any property
* **Transcript Analysis** — Deep reading of call transcripts with speaker attribution
* **People & Company Lookup** — Find and cross-reference contact and organization data
* **Document Reading** — Analyze the full content of any document in your workspace
* **Pattern Detection** — Identify trends, anomalies, and recurring themes

<Info>
  The agent decides which tools to use based on your question. You don't need to specify how it should research—just ask what you want to know.
</Info>

## Getting Started

<Steps>
  <Step title="Open Chat">
    Navigate to **Chat** in the left sidebar or start a new chat from the dashboard
  </Step>

  <Step title="Ask a Research Question">
    Ask something that requires investigation—the more specific, the better the results. For example: "What are the biggest pain points our enterprise customers raised this month?"
  </Step>

  <Step title="Watch the Agent Work">
    See which tools the agent uses, what data it queries, and how it builds its analysis in real-time
  </Step>

  <Step title="Follow the Evidence">
    Click any citation to jump to the original call timestamp, signal, or document. Verify findings and explore further.
  </Step>

  <Step title="Ask Follow-Ups">
    Refine your analysis with follow-up questions. The agent maintains full context from your conversation.
  </Step>
</Steps>

## Tips for Better Results

<Check>
  **Be specific about what you want to know**: "What pricing objections came up in enterprise deals last quarter?" beats "Tell me about pricing"
</Check>

<Check>
  **Scope when you can**: If you're researching a specific customer, mention them. If it's about a product area, name it.
</Check>

<Check>
  **Ask follow-ups**: The agent remembers your full conversation. Build on previous answers to go deeper.
</Check>

<Check>
  **Use it for prep**: Before a customer call, ask the agent to summarize everything that customer has discussed, flagged, or requested.
</Check>

<Check>
  **Verify with citations**: Click through to source material when making important decisions. The agent is powerful but always check the evidence.
</Check>

## From Analysis to Action

Agent Chat responses aren't just for reading—you can act on them:

* **Generate Documents**: Turn any analysis into a formal document (PRD, brief, report)
* **Generate Highlight Reels**: Create video compilations from referenced call moments
* **Add to Collections**: Save referenced items to folders for your team
* **Manage Projects and Inbox Items**: Create, update, merge, promote, or attach evidence with confirmation for destructive actions
* **Share Insights**: Copy responses with properly formatted links for teammates

<Info>
  Agent Chat is currently in beta and available to workspaces with the feature enabled. Contact your workspace admin or support to get access.
</Info>