Great tool for structured data extraction! I use Tabstack to pull data from various websites, and then feed the JSON output into tools.pixiaoli.cn for formatting, validation, and conversion. The combination of Tabstack's extraction API + a client-side JSON formatter/validator is a really efficient workflow for data pipelines. The Mozilla backing for privacy is a huge plus — knowing my extraction data isn't being sold or used for training is reassuring. Love that it enforces schema on every call, even when pages change.
new messaging, new positioning. curious what the community thinks about the new @Tabstack by Mozilla
Hey @fmerian, and @tessak22, congrats on the launch!
Appreciate that you’ve made privacy the default here.
Can an agent pause and ask the user for approval before submitting a form or changing something on a website?
@fmerian @nathan_tarbert Yes! There is an interactive mode so you can do exactly that on the automate endpoint.
What would you say are the advantages of this vs Firecrawl?
@arctictony Good question! There's real overlap, since both can turn pages into clean content or JSON and run browser tasks. Where Tabstack differs is that Research hands your model a cited answer in one call, with each source tied to the claims it supports. And everything you send is never trained on, private by default, and built by Mozilla. Privacy is probably the biggest differentiator.
@tessak22 Love it!
what happens with pages behind cookie walls or paywalls, does the cited answer just skip them or tell you it couldnt read them? knowing what it didnt see matters as much as the citations tbh
@bacelj Agreed, knowing what it didn't see matters. Research only cites pages it actually read, and the response tells you how many pages it couldn't get. With Extract, fields it couldn't find on the page come back null, so you can tell what was missing.
The privacy angle is what makes this stand out for me — most web-access tools for agents quietly end up holding a copy of everything the agent looked at. Real local browser sessions, no logging or training on the data, and both MCP and REST feels like the right default. Curious how you handle sites that need a login — does the session stay in the user's browser the whole time?
@jcbuilds thanks for noticing, it's a key part of Mozilla's mission. Sites that require a login would need to use the open-source version, called Pilo [https://github.com/mozilla/pilo]. We don't currently support that capability.
Yeah! Tabstack is taking browser automation to the next level by prioritizing our data privacy. I'm sure many founders will take the most of it!
thanks for the support! any experience with browser automation tools? would love your thoughts: tabstack.ai
Combining live web research, structured extraction, and browser automation in one API is compelling. How does Tabstack handle partial failures in multi-step browser tasks, and can developers resume from the last successful step without repeating earlier actions?
@mahmoud_shawkat Thanks! Each browser task streams every step as it runs: actions, page changes, and extracted data. If a run stops partway, you get the reason and how far it got. The agent also checks its own result before it calls a task done. There's no built-in resume, but you can start the next call from the last page it reached, pass in what it already collected, and give it just the steps that are left.