> For the complete documentation index, see [llms.txt](https://docs.upriverdata.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.upriverdata.com/product-usage/agent.md).

# Agent

Chat with an AI agent that investigates, queries and fixes your data.

Upriver's Agent turns your entire data stack into something you can simply ask questions of. It's an AI-powered chat workspace built directly on top of the context Upriver already understands about your data — your schemas, lineage, pipelines, incidents and history — so instead of manually digging through tables, dashboards and code, you just describe what you want to know or fix, and the agent takes it from there.

<figure><img src="https://875415170-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FLE0zsJmUpeLjYw085cnO%2Fuploads%2Fgit-blob-0265cf23db59214127d5881e35ade1ecd6645f20%2Fimage%20(agent-start).png?alt=media" alt=""><figcaption><p>The Agent landing page, with a chat input and suggested starting points</p></figcaption></figure>

## Starting a conversation

The agent page opens with a chat input and a row of suggestion chips to help you get started, for example:

* "Why did rows drop?" — investigates why a specific model dropped rows and looks for a root cause across upstream datasets, joins and recent pipeline changes.
* "Why did this model break?"
* "What changed in this pipeline?"
* "What does this model affect?"
* "Write SQL" — asks the agent to write a SQL query to investigate a specific data issue.

If your workspace doesn't yet have a data or code integration connected, a special chip prompts you to set one up first.

You can also just type a free-form request in the chat box — press `Enter` to send, `Shift + Enter` for a new line. While the agent is working, the send button turns into a **Cancel** button so you can stop it at any point.

## What the agent can do

Because the agent is grounded in the context Upriver already generates — schemas, profiles, lineage, past incidents and pipeline history — it doesn't just guess at answers, it reasons from a real, continuously up-to-date model of your data. That's what lets it move fast **and** get it right: building validated pipelines and queries, and answering questions about your data stack with the confidence of someone who already knows how everything connects.

As it works, the agent streams its progress back into the chat. This can include:

* A step-by-step **task plan** it is executing, shown as a checklist.
* Tool calls, such as running a SQL query, reading or editing a file, or running a command.
* Notifications when it creates or edits a file.
* Questions or confirmations it needs from you before continuing.
* A final **summary** once it has finished the current piece of work.

Everything the agent produces along the way — code, catalog lookups, SQL queries and their results — is organized into a drawer on the side of the chat, described below.

## Tasks

Every conversation with the agent is tracked as a **task** in a collapsible sidebar. From the task list you can:

* Start a new task or switch between existing ones.
* Rename or delete a task.
* Share a task, generating a link others can use to view the conversation.

If your workspace is connected to a git repository, the drawer header also shows a branch switcher (admin only), letting you point the agent at a specific branch before starting a task. `Git push`/`pull` buttons and a workspace download button let you sync or export the agent's working files.

## The helper drawer

The panel to the right of the chat — the helper drawer — is where the agent surfaces everything it touches: code changes, data catalog entries, query results and a full SQL editor. It's organized as tabs, and you can open several tabs side by side by adding a split panel (the columns icon in the drawer toolbar).

<figure><img src="https://875415170-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FLE0zsJmUpeLjYw085cnO%2Fuploads%2Fgit-blob-9504957f996002e655186d146767615724830d50%2Fimage%20(agent-drawer).png?alt=media" alt=""><figcaption><p>The helper drawer, showing the agent's work alongside the chat</p></figcaption></figure>

Each open tab has a "..." menu with actions such as Copy, Save, Duplicate, Rename and Close, and a search popover lets you open a new file, table or query tab directly. The drawer itself can be resized by dragging its left edge.

The main things you'll find in the drawer are:

* [Code](/product-usage/agent/code.md) — the files the agent read or edited.
* [Catalog](/product-usage/agent/catalog.md) — data assets the agent looked up, including sample data.
* [Columns and Lineage](/product-usage/agent/columns-and-lineage.md) — column-level detail and upstream/downstream lineage for any table.
* [Queries and Charts](/product-usage/agent/queries-and-charts.md) — every warehouse query the agent ran, with results and charts.
* [Query Editor](/product-usage/agent/query-editor.md) — a full SQL editor to run, save and re-run queries yourself.

Agent conversations can also be automated on a schedule — see [Routines](/product-usage/routines.md).