> ## Documentation Index > Fetch the complete documentation index at: https://docs.ingestly.ai/llms.txt > Use this file to discover all available pages before exploring further. # Assistant > Build workflows, debug runs, and find your way around Ingestly by chatting with the in-app assistant. The assistant is a chat panel that knows the page you are on. In the workflow editor it proposes changes to your workflow. On a run it explains what happened. Anywhere else it answers questions about Ingestly from this documentation. The assistant is enabled per organization. If you do not see the **Open Assistant** button in the header, it is not turned on for your organization yet. Contact [support](mailto:support@ingestly.ai) to request it. ## Open the assistant * Click **Open Assistant** in the app header or the workflow editor toolbar. * Press **Ctrl+J** (**Cmd+J** on Mac) on any page. * Open the command palette with **Ctrl+K** and choose **Open Assistant**. The panel can be resized. Its header holds **Chat history**, **New chat**, and **Close assistant**. If you move to another page mid-chat, the conversation continues and a divider shows what you are now viewing. ## In the workflow editor Ask the assistant to build or change the workflow you have open: add nodes, connect them, configure them, rename or remove them, or mark extraction fields as required. It can also read the current graph, including unsaved edits, and check whether the workflow is ready to run. The assistant never changes your workflow on its own. Each change arrives as a **proposal**: * The proposal lists every change, with a status line such as **Proposed 3 · Applied 0 · Skipped 0**. * Expand a change with **Show what changed** to see a before and after view. * Click **Skip this change** to leave one out. * Click **Apply all** (or **Apply selected** when you skipped some) to apply them to the canvas, or **Skip all** to discard them. Applied changes land on the canvas like your own edits. Save the workflow to keep them. To stop approving every proposal in a long session, tick **Auto-apply the rest of this chat's changes**. It lasts for the current chat only and is off again after a reload. Proposals cannot be applied when the workflow is [locked](/workflows/locking) or you lack permission to edit it. The assistant can still answer questions about it. ## On a run Open a run and ask why it failed, what a step received or produced, or whether a step's configuration changed since the run. The assistant reads step details, the run's output, and sub-runs. It can also run a step again or test a node against the run's document. Both cost credits, so the assistant always asks first, and you choose **Run it** or **Don't run**. Auto-apply never skips this question. It can also open the workflow in the editor with the relevant node selected. ## In the learning pane In a workflow's **Learning** tab, ask the assistant to explain a learned instruction and the reviewer corrections behind it. See [self-learning](/workflows/self-learning). ## What the assistant sees The assistant works from the page you are on and from this documentation. Before your workflow is sent to it, HTTP Request header values that look like credentials are replaced with `[redacted]`. That covers authorization and cookie headers, any header whose name contains key, token, secret, signature or password, and values that start with Bearer, Basic, Digest or Token. Chat history is kept in your browser only, separately for each signed-in user. The most recent 10 chats are kept. History does not follow you to another browser or device. ## Permissions and limits | What | Requirement | | - | - | | Chat | Permission to view workflows | | Run a step again, test a node | Permission to create runs | | Apply a proposal | Permission to edit the workflow, and the workflow is not locked | Each organization can send up to **60** assistant requests per hour and **300** per day. Past the limit, the assistant asks you to wait and try again. A very long chat or a very large workflow can exceed what the assistant can read. It tells you when that happens. Start a new chat to continue. ## Cost The assistant draws on your organization's credits, billed by usage rather than by page. A short question costs a small fraction of a credit, and a long session with several steps costs more. At zero credits the assistant stops answering until credits are added. See [assistant chat](/admin/credits#assistant-chat) in credits. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.