> ## Documentation Index > Fetch the complete documentation index at: https://docs.valkyrieapp.azumo.com/llms.txt > Use this file to discover all available pages before exploring further. # AI assistant > Fine-tune by chatting: attach a spreadsheet or text, say what you need, and confirm. The AI assistant is a chat on [Home](/dashboard/overview#home) that works through Valkyrie for you. Tell it what the model should get better at, attach your data if you have it, and it builds the training data, picks a base model, prices the run and prepares the fine-tune. **Nothing that costs money happens until you confirm.** Valkyrie AI assistant conversation: the steps it took and a Start fine-tune card with time, cost and balance ## Start a conversation Make sure **Do it with the AI assistant** is selected at the top of the box. Write it in your own words, for example "Train a model that predicts which customers will churn" or "Answer our customers the way our support team does". Use **Upload a file**, **Paste text**, or **Use a dataset**. You can attach more on any later message with the paperclip. Click **Ask the assistant**. Each step it takes shows as a short line in the chat, with a link to anything it created. Valkyrie Home start box in AI assistant mode with Paste text, Upload a file and Use a dataset ## What you can attach Files can be up to 200 MB: * **Tables**: Excel (.xlsx, .xls), CSV, TSV, JSON, JSONL and Parquet. A table is saved as one of your [datasets](/dashboard/data). * **Text**: TXT and Markdown, such as an FAQ, policies or past replies. You can also start from one of your datasets or an [example set](/dashboard/data#start-from-an-example-set), or attach nothing and describe the task. ## What it can do * **Turn a table into training data**: attach a table and say which column to predict, for example "predict whether a customer churns". The assistant converts every row exactly, leaves out columns like IDs and names, and uses the column's values as labels (or as answers, for numbers and free text). * **Build a dataset from text**: from an FAQ or policies, it writes question and answer examples using only facts from your material. This runs in the background for a few minutes. * **Pick a base model** from the ones Valkyrie has tested, and say why. * **Tell you the time and cost** of a fine-tune, and your wallet balance. * **Start the fine-tune**, then tell you how it is going and, when it is done, how much better it got against the base model. See [Test results](/dashboard/models#test-results). * **Deploy it** as an endpoint, running until you stop it or on office hours. ## You stay in control Before a fine-tune or a deploy, the assistant shows a card with the model, the data, the time, the cost and your balance. It only starts when you click **Start fine-tune** or **Deploy**, and **Not now** cancels it. Your conversation stays in the page address, so you can leave and come back, and **Pick up where you left off** under the box lists your recent conversations. **New conversation** starts over. ## Prefer to do it yourself? Switch Home to **I have my examples**, or open [New fine-tune](/dashboard/fine-tuning) from the top bar. The [Fine-tune a model](/guides/fine-tuning) guide covers both ways. Where the assistant isn't available yet, Home opens on **I have my examples**, and **Do it with the AI assistant** shows a padlock with a link back to your own examples. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.