---
title: Build with Tabstack | Tabstack
description: Start with a current answer, a URL, or a public web task. Find the quickstart, SDK, integration, API reference, schema guide, and troubleshooting path for the job you are building.
---

Start with a current answer, a URL, or a public web task. Find the quickstart, SDK, integration, API reference, schema guide, and troubleshooting path for the job you are building.

## What Tabstack is

Tabstack is a managed web API for people and teams running their own models. Ask a question, pass a URL and schema, or describe a public web task. Tabstack handles the web workflow inside the call and returns a cited answer, clean text, matching JSON, or the completed result.

Search is access to the web. Tabstack handles the work that comes after search: choosing sources, fetching pages, recovering readable content, identifying gaps, searching again, reconciling disagreement, and attaching citations. That workflow runs inside the call instead of inside your model.

## Ask for the output you need

| You send                | You get back                     | Endpoint            |
| ----------------------- | -------------------------------- | ------------------- |
| A question              | A cited answer from live sources | `/research`         |
| A URL                   | Clean text                       | `/extract/markdown` |
| A URL and a JSON schema | Matching JSON                    | `/extract/json`     |
| A URL and instructions  | Matching JSON                    | `/generate/json`    |
| A public website task   | The completed task               | `/automate`         |

## Which call should I make first

`/research`. It is the call that shows the difference between search results and a finished result: you send a question and get back an answer with the sources it cited. The [Quickstart](/getting-started/quick-start/index.md) takes you from API key to that response.

From there, [Extract](/guides/how-to-use-markdown-endpoint/index.md) is the call you will make most often in a pipeline, and [Automate](/guides/how-to-use-automate-endpoint/index.md) handles the cases where reading a page is not enough.

Make a research call

[Authenticate and get a cited answer](/getting-started/quick-start/index.md), starting with the cheap setup check that proves your key works.

Read the Research guide

[Modes, the progress stream, citations, and timeouts](/guides/research/index.md) for the endpoint that leads.

Install for your stack

The [Python SDK](/sdks/python/quickstart/index.md), a [CLI](/getting-started/cli/index.md) for shell and agent workflows, the [Hermes plugin](/integrations/hermes/index.md), and the [TypeScript SDK](/sdks/typescript/quickstart/index.md).

Write a schema

[Schema design](/guides/schema-design/index.md) is what decides whether matching JSON matches what you actually wanted.

## Trust

Never trained on. Private by default. Built by Mozilla.

See [how data is handled](/trust/data-handling/index.md) for what each endpoint processes, what is retained, and where hosted processing ends and self-hosted [Pilo](/guides/pilo/index.md) begins.

For detailed endpoint documentation, see the [API Reference](/api/index.md).
