Skip to content

Quickstart

This page takes you from nothing to your first structured output in about five minutes. We'll extract a product name and price from a tiny CSV.

By the end you'll have created a dataset, an output schema, and a task — and run it.

Two ways to follow along

Every step below can be done in the Studio (no code) or by hand (plain files). We show both. Pick whichever you prefer — they produce identical files.

0. Before you start

Make sure you've installed the package and that Ollama is running. You don't need to pull a model in advance; the tool downloads the one you request on first use.

Create a working folder with the standard layout:

mkdir -p myproject/{data,tasks/parsers,examples,output}
cd myproject

The tool looks for data/, tasks/, tasks/parsers/, and writes to output/ by default.

1. Add a dataset

Save this as data/products.csv:

id,text
1,"A 250 ml bottle of extra-virgin olive oil, on sale for €4.99."
2,"Stainless steel water bottle (750ml) — now only 18.50 euro."

The important part is that there's a column with the text to read — here, text. See Preparing data for JSON input and other details.

2. Define the output schema

The schema is a Pydantic model whose top-level class is named OutputParser.

Launch the Studio (launch-extractinator), go to the Task tab, choose Build a new task, and click 🛠️ Build new next to Output schema. Add two fields — product_name (str) and price (float) — then Save & use this schema. The Studio writes the file into tasks/parsers/ for you.

Save this as tasks/parsers/product_schema.py:

from pydantic import BaseModel

class OutputParser(BaseModel):
    product_name: str
    price: float

See Output schema for optional fields, lists, and nested models.

3. Create the task

The task JSON ties the dataset, the text column, and the schema together.

Still on the Task tab under Build a new task: pick products.csv as the dataset, choose text as the text column, make sure product_schema.py is selected, write a short description, and click 💾 Save task. The task is now marked ready.

Save this as tasks/Task001.json:

{
  "Description": "Extract the product name and price from each row of text.",
  "Data_Path": "products.csv",
  "Input_Field": "text",
  "Parser_Format": "product_schema.py"
}

4. Run it

Open the Run tab, pick a model (e.g. phi4), and click 🚀 Run. Watch the log stream; when it finishes you'll get a success summary and a nudge to open Results.

extractinate --task_id 1 --model_name "phi4"

The --task_id 1 matches Task001.json. On first run, Ollama downloads phi4 if you don't have it yet.

5. Look at the results

Results land in:

output/run/Task001-run0/nlp-predictions-dataset.json

Each record has your original columns (id, text), the extracted fields (product_name, price), and a status:

[
  {
    "id": 1,
    "text": "A 250 ml bottle of extra-virgin olive oil, on sale for €4.99.",
    "product_name": "Extra-virgin olive oil 250 ml",
    "price": 4.99,
    "status": "success"
  }
]

In the Studio, the Results tab shows the same data with success/failure counts, a searchable table, and a per-record detail view. More on the format in Understanding output.

Where to go next