Tabstack by Mozilla
Extract web data and automate browsers, no scraper required.
Start new thread

Tabstack Web Research - Run a research agent with cited answers in a single API call

by•
/research gives your app or agent cited answers from the live web in one API call. Not a pre-indexed corpus: the actual live web. Every request comes back with source URL's users can verify. Source selection, synthesis, and citation formatting are all inside the call. You write or maintain none of that code. Built for legal, financial, and competitive intel, where a wrong answer is a liability. Free to try.

Add a comment

Replies

Best
Maker
📌

We launched /research today, a single API call that runs a full web research agent and returns cited answers.

Here's the pattern it replaces:

// What most teams build:

// 1. Search the web (or an index)

// 2. Fetch and read relevant pages

// 3. Synthesize across sources

// 4. Validate citations

// 5. Format for streaming

// 6. Debug all of the above forever


// What /research does:

const stream = await client.agent.research({ query: 'Your question here' })

That's it. The orchestration lives inside the call. Your codebase doesn't own it.

First 10,000 credits are free. Happy to answer questions in the comments. Give it a try →


Three things that matter about how we built this:

1. Citations are not optional.

Every claim comes back with a source URL. Users can verify where the answer came from. That's the difference between an AI feature people trust and one they don't.

2. Live web, not an index.

Pre-indexed research APIs (Exa, Perplexity) are fast and good for broad questions. /research goes to the live web on every call. For legal, financial, and competitive intelligence work — where the answer needs to be current and sourced — that's the relevant difference.

3. SSE streaming built in.

Users watch it work in real time. No spinning wheel. Progress events as sources are found, then a complete event with the full report and all citations.

Maker

 S/O for this new launch! Psyched to be part of this journey

Maker

 We're thankful for your part in the journey!

Maker

just the beginning

Maker

happy to share this new collaboration with and the team!

First launched on last April, the -backed platform is launching again today, introducing a new /research endpoint.

Run agents to explore the web and answer complex questions with precision, all in a single API call. Simple, accurate, and secure. There are multiple products in the browser automation category - just hits differently.

Read the docs to learn more - - and start for free here:

Maker

 So thankful for you and the support! It's been a ton of fun bringing this to life with you!

Cited or it didn’t happen. Single API calls make it dangerous.
Maker

Cited or it didn’t happen. Single API calls make it dangerous.

Spot on!

Maker

quick update: we're launching a new /automate endpoint today. join the launch:

Maker

 exactly. I wish all LLM's had to cite their sources. It would make results a lot more trustworthy. I actually use Tabstack for almost everything I do with LLM's for this reason.

How do you pick sources and avoid low quality SEO pages, and can users whitelist domains?

Maker

 Great question, and TL,DR: The system prioritizes official docs, ignores SEO-optimized posts, and filters out marketing noise and DOM boilerplate during content extraction. That's it!

Give it a spin at - curious to have your feedback about the onboarding flow and overall DX

Maker

 quick update: just launched a set of tools that integrate with your current dev workflow: an MCP server, CLI, Raycast extension, and an Agent Skill.

Maker

 We don't support a domain whitelist, but I'm submitting that as a possible feature request. Thanks for your engagement and support.

Liked the no-scraper web-extract angle — gave it a quick test with Lastest, run here:

Maker

 what led you to a 404? I'll get it addressed ASAP.

The citations part is what matters most to me here. For research workflows, a good looking answer is not enough if people cannot check where it came from. Live web also makes sense for topics that change quickly, especially competitive research or financial updates. How do you handle cases where sources disagree with each other? Do you show conflicting findings or choose the strongest source?

Maker

The citations part is what matters most to me here. For research workflows, a good looking answer is not enough if people cannot check where it came from. Live web also makes sense for topics that change quickly, especially competitive research or financial updates.

Exactly. From my perspective, it's all about trust - especially these days - and has an unfair advantage here, backed by . Read their manifesto here:

re: conflict resolution.

How do you handle cases where sources disagree with each other? Do you show conflicting findings or choose the strongest source?

Good question! If the loop encounters conflicting information, the system resolves this by verifying the context. The result is a reconciled data point that prioritizes the most authoritative source. Learn more in this blog announcement:

Hope this helps!

The “citations are not optional” stance is exactly right, especially for legal / financial / competitive-intel use cases.

One thing I’d be curious about: do you treat citations as source links only, or as claim-level support? In research workflows, a source URL is helpful, but the trust really jumps when the user can see which sentence or assertion each source is meant to support, plus whether another source disagreed or was skipped as stale.

That extra bit of provenance would make the output much easier to use in a report without forcing someone to re-run the whole research process manually.

Awesome to see this launching!

The "cited answers from the live web, source URLs users can verify" point is exactly the bar for any signal you'd act on financially. I work in finance and also built PolyMind (polyminds.netlify.app), which pulls real-time signal from Polymarket trades — and the hardest part was never fetching data, it was making every alert traceable back to a verifiable source so a user can sanity-check it before trusting it. Curious how Tabstack handles conflicting sources on the same question — does /research surface the disagreement, or pick a winner?

Maker

Curious how Tabstack handles conflicting sources on the same question — does /research surface the disagreement, or pick a winner?

That's the best part of 's Autonomous Research. The system does not just aggregate data, it reconciles it. The loop might encounter conflicting information, and the system resolves this by verifying the context. The result is a reconciled data point that prioritizes the most authoritative source.

Read the docs for more technical details:

PolyMind sounds awesome! and ngl a perfect use case for . Curious what's your current stack?

The live-web approach vs pre-indexed APIs is a real tradeoff I've thought about while building research features. Pre-indexed is faster and cheaper at scale, but for anything time-sensitive — earnings reports, product launches, competitive moves — the index lag can be 24-48h which is basically useless. Live web is the only option there.

One thing I'm curious about: when sources conflict, the "prioritize most authoritative source" approach makes sense, but how do you define authoritative? Domain reputation, recency, citation count? I've seen cases where a newer but less-established source had the correct info and an older authoritative source was stale. Would love to see the conflict resolution logic exposed more transparently in the output.

12
Next
Last