> ## Documentation Index
> Fetch the complete documentation index at: https://docs.getcollate.io/llms.txt
> Use this file to discover all available pages before exploring further.
# Collate AI Chat
> Ask questions about your data in plain language and get back charts, insights, and answers — powered by Collate's knowledge graph.
# Collate AI Chat
Collate AI Chat is the conversational AI interface inside Collate. Ask questions about your data in plain language and Collate AI returns charts, insights, and answers — without writing SQL or knowing where the data lives.
Collate AI Chat runs with your permissions, so every action it takes — querying data, editing metadata, creating tests — respects your access level.
**Note**: When Collate AI Chat queries your connected data sources, it uses read-only SQL operations such as `SELECT`, `WITH`, `SHOW`, `DESCRIBE`, and `EXPLAIN`. Depending on your permissions, Collate AI Chat can still create or update metadata inside Collate, such as glossary terms, tags, owners, and data quality tests.
## Core Capabilities
Collate AI Chat connects directly to your organization's knowledge graph — the layer built from your metric definitions, glossary terms, lineage, table schemas, and profiling data. This gives it accurate, organization-specific answers rather than generic approximations.
Here's what happens when you ask a question:
1. **It finds the right definition first.** If you ask about customer lifetime value, Collate AI Chat looks up your organization's agreed-upon formula from the business glossary — not a generic approximation. Every teammate who asks the same question gets the same canonical answer.
2. **It discovers the right tables and joins.** Using lineage and schema metadata, Collate AI Chat figures out which tables to query and how to join them — including intermediate tables that aren't obvious from the question.
3. **It filters using real data.** Column value filters come from profiler agent execution results — not guesses — so results reflect what's actually in your data.
4. **It shows its work.** SQL queries appear inline as Collate AI Chat works through your question. You can see exactly what ran, which tables were used, and which definitions were applied.
## What You Can Do
* **Answer business questions**: Ask in plain language — Collate AI Chat finds the right tables, joins, and definitions, then runs the query.
* **Generate and show SQL**: Every answer shows the SQL that ran, the tables used, and the definitions applied — so you can verify or hand it off to your data team.
* **Build and refine charts**: Bar, line, pie, donut, number, scatter, heatmap, and table charts — generated from your data and refined with follow-up prompts.
* **Pin to dashboards**: Save any chart to an AI Analytics dashboard for recurring use, scheduled refresh, and team delivery.
* **Share conversations**: Invite teammates as EDITOR or VIEWER so they see the same charts and reasoning and can continue the analysis together.
* **Schedule reports**: Deliver charts or dashboards via email or Slack on a recurring cadence.
* **Explore assets**: Browse and discover data assets from within Collate AI Chat without leaving the interface.
* **Manage metadata**: Create or edit glossary terms, assign owners, tag assets, and create data quality tests — from the chat.
* **Manage your profile**: Access your personal access tokens and connection settings directly from your Collate AI Chat profile.
## Prerequisites
* Collate AI is enabled in your workspace. Contact your admin if you are unsure.
* Metadata, lineage, and usage ingestion jobs are running so Collate AI Chat has context.
* You have permissions to view and act on the assets you are asking about.
## Use Cases
### Data Analyst: Revenue Analysis
Ask a business question and Collate AI Chat finds the relevant certified tables, generates the SQL, and returns results with a chart.
* **Prompt**: "Give me total market share for eco vs non-eco products by revenue. Only Tier 1 data."
* **Result**: Finds certified Snowflake tables (`products`, `orders`, `customers`), generates a join query, and returns the revenue split with a chart.
* **Follow-up**: "Use a donut chart with dark green for eco." — Collate AI Chat updates the chart style and summary.
### Data Analyst: Customer Segmentation
* **Prompt**: "What percent of customers bought eco products, grouped by customer tier?"
* **Result**: Produces SQL, groups by tier, and surfaces the insight that higher-tier customers engage more with eco products.
### Data Steward: Glossary and Tagging
* **Prompt**: "Do we have a definition for revenue?"
* **Result**: Searches glossary terms and suggests creating one if missing.
* **Prompt**: "Create a term for revenue calculation and tag relevant Snowflake tables."
* **Result**: Adds the term, links it to assets such as `order_items`, and records ownership.
### Data Engineer: Lineage and Data Quality
* **Prompt**: "Show lineage for the customer report dashboard."
* **Result**: Maps upstream tables (including `customer_360`) and downstream impacts.
* **Prompt**: "Search for anomalies in these tables: nulls, negatives, duplicates."
* **Result**: Surfaces outliers such as negative `unit_price` values.
* **Prompt**: "Create test cases for the products table."
* **Result**: Suggests completeness, uniqueness, range, and integrity tests and records who requested them.
## Working with Charts
Ask a question and Collate AI returns results as charts automatically — no SQL, no BI tool, no configuration needed. Charts appear in the conversation as Collate AI works through your question.
### Explore a Chart
Each chart has three views:
| View | What it shows |
| --------- | ------------------------------------------------------------------------------------------- |
| **Chart** | The default view. Adjust axes, groupings, and chart type from the settings on the card. |
| **Table** | The same data as rows and columns — spot-check numbers, sort, or copy data out. |
| **Code** | The SQL query that produced the result — verify the logic or run it in your own query tool. |
Use the **Chart Selector** dropdown when Collate AI returns multiple charts for a complex question.
### Customize a Chart
Use follow-up prompts to refine charts without starting a new conversation:
* "Use a donut chart instead."
* "Break this down by region."
* "Only show the top 10 results."
* "Add a filter for Tier 1 data only."
### Pin to a Dashboard
Save useful charts to an AI Analytics dashboard for recurring use, scheduled refresh, and team delivery.
1. On any chart in the conversation, click **Pin to dashboard**.
2. Choose an existing dashboard or create a new one. Use the search box to find a dashboard by name quickly.
3. Open the dashboard from the **AI Analytics** page to see all your pinned charts.
Keep pinning charts from any conversation to build a complete picture over time. See [AI Analytics](/ai-2-0/collate-ai/ai-analytics) to manage dashboards, schedule refreshes, and deliver reports to your team.
## Share a Conversation
You can share any Collate AI Chat conversation with teammates — they open it and see the exact same charts, SQL, and reasoning you saw, and can keep asking questions right where you left off.
**To share a conversation:**
1. Open the conversation you want to share.
2. Click **Share** on the conversation.
3. Enter the email addresses or user groups, set their role — **Editor** (can ask follow-up questions) or **Viewer** (read-only) — and click **Invite**.
4. Your teammate opens the shared chat and sees the same charts, SQL queries, and reasoning — not just a screenshot.
**What teammates see when they open a shared chat:**
* Every chart you generated, in the same order.
* The SQL query behind each chart — so they can verify the logic themselves.
* The AI's reasoning steps — so they understand how the answer was derived.
* A live input to continue the analysis with their own questions (Editors only).
This applies to any conversation — not just ones with charts. Share an asset exploration, a governance workflow, or a data quality investigation just as easily.
## Best Practices
* Be specific — include time ranges, owners, schemas, or tier levels to narrow results.
* Use follow-up prompts to refine charts rather than starting a new conversation.
* Review the SQL Collate AI Chat shows to confirm the query matches your intent before acting on the result.
* Pin useful charts to a dashboard and schedule a refresh so the numbers stay current without re-asking.
* Share conversations with teammates when collaborating so everyone sees the same charts and reasoning.
* Paste failing SQL or error context when asking for fixes so Collate AI Chat has what it needs to help.
## Troubleshooting
* If answers seem incomplete, verify that recent metadata and lineage ingestion runs succeeded.
* If you cannot see certain assets, confirm your Collate permissions for those services.
* Collate AI responds in your UI language setting, not necessarily the language you typed your question in. To change it, click the **Profile** icon at the bottom of the left sidebar, then select **Language**.
* For inconsistent results, restate the question with the relevant schema or service name.
* For deeper issues — wrong formulas, wrong table surfacing, filters returning zero rows — see [Improving AI Answers](/ai-2-0/collate-ai/context-engineering#improving-ai-answers) for catalog-level fixes.