Research

Get a cited answer from live sources in one call.

Tabstack plans the search, reads the sources, checks for gaps, synthesizes the answer, and attaches citations before the response reaches your model or application.

Make a research call Read the research guide

Progress stream · Multi-source answer · Source citations

"which runtimes shipped web search?"

ollama.com/blog

docs.openwebui.com

ANSWER

Two runtimes shipped it.

ollama.com/blogdocs.openwebui.com2 cited

The Research API

What is the Tabstack Research API?

Tabstack Research takes a question, searches and reads live sources, reconciles what it finds, and returns a cited answer in one API call. Your model receives the result instead of operating the search and fetch loop itself.

one question

post /research

query: "which runtimes…"
mode: "balanced"

one cited answer

The problem after search

Search gives your model sources. It still has to build the answer.

Building the answer means selecting and reading pages, checking what is missing, searching again, reconciling conflicting information, synthesizing the result, and attaching citations. A search endpoint can be the right choice when your model reliably completes those steps. Tabstack Research completes them inside the call and returns the cited answer as a usable artifact.

Step	Search endpoint	Tabstack Research
Find relevant pages	Returns results or snippets	Handled inside the call
Retrieve and read pages	Your stack or model	Handled inside the call
Identify missing information	Your model	Handled inside the call
Reconcile sources	Your model	Handled inside the call
Build the answer	Your model	Returned by the call
Attach citations	Your stack	Returned with the answer

Inside the call

Tabstack builds the answer before the response comes back.

Interpret the question

Work out what evidence the answer needs.

Find and read sources

Read current pages, not training data.

Fill the remaining gaps

Search again when the first sources fall short.

Synthesize one answer

Build a single answer across all the evidence.

Cite the sources

Each answer comes back with the pages that support it.

research.tsTypeScript

import Tabstack from '@tabstack/sdk'

const client = new Tabstack()

const stream = await client.agent.research({
  query: 'Which runtimes shipped web search this month?',
})

for await (const event of stream) {
  if (event.event === 'complete') {
    console.log(event.data.report)
    console.log(event.data.metadata.citedPages)
  }
}

from tabstack import Tabstack

client = Tabstack()

for event in client.agent.research(
    query='Which runtimes shipped web search this month?',
):
    if event.event == 'complete':
        print(event.data.report)
        print(event.data.metadata.cited_pages)

tabstack agent research 'Which runtimes shipped web search this month?'

The response

Inspect the request, progress, answer, and sources.

Planning the search
Reading sources
Checking for gaps
Building the answer
Attaching citations

Note: Your model does not need to select the sources or carry raw page content through its own context window.

Use cases

Use research when a question should end in an answer.

Answers in your own assistant

Give an own-model assistant sourced information past its cutoff.

Claim checking at draft time

Check a draft answer against current sources before it ships.

Current documentation lookup

Answer an implementation question from current technical sources.

Research inside your product

Add a research action that hands your user a cited result.

Briefings on a subject you define

Research the subject and pass the result to your workflow.

Choosing the call

When another call fits better.

Already know the page? Extract returns its content or matching JSON.
Need something done on a public site? Automate runs the task and returns the result.
Only need results and snippets? A plain search API is the lighter call.
Is the loop itself your product? Build it yourself.

Controls and limits

Choose the research depth your job needs.

Research bills per action. A call runs several actions; the count depends on the question. A Fast action costs 250 credits and a Balanced action 350; Balanced is available on Team and Pro.

Mode	Credits
Fast	250
Balanced	350

Trust

How Research handles your data.

Never trained on. Private by default. The Trust page explains what is processed and the account controls.

How data is handled

your query

tabstack

reads public pages
builds the answer

never trained on

private by default