Tabstack – Browser infrastructure for AI agents (by Mozilla)
Hi HN, My team and I are building Tabstack to handle the "web layer" for AI agents. Launch Post: https://tabstack.ai/blog/intro-browsing-infrastructure-ai-ag... Maintaining a complex infrastructure stack for web browsing is one of the biggest bottlenecks in building reliable agents. You start with a simple fetch, but quickly end up managing a complex stack of proxies, handling client-side hydration, and debugging brittle selectors. and writing custom parsing logic for every site. Tabstack is an API that abstracts that infrastructure. You send a URL and an intent; we…
In plain words
Tabstack is an API that handles web browsing infrastructure for AI agents. Instead of developers managing complex stacks of proxies, rendering, and custom parsing for each website, they send Tabstack a URL and an intent, and receive clean, structured data back. The service uses escalation logic to start with lightweight fetches and only deploy full browser automation when needed, reducing costs and complexity.
written from the facts on this page · September 2026
From the sources
In the maker’s words, at launch
Hi HN, My team and I are building Tabstack to handle the "web layer" for AI agents. Launch Post: https://tabstack.ai/blog/intro-browsing-infrastructure-ai-ag... Maintaining a complex infrastructure stack for web browsing is one of the biggest bottlenecks in building reliable agents. You start with a simple fetch, but quickly end up managing a complex stack of proxies, handling client-side hydration, and debugging brittle selectors. and writing custom parsing logic for every site. Tabstack is an API that abstracts that infrastructure. You send a URL and an intent; we handle the rendering and return clean, structured data for the LLM. How it works under the hood: - Escalation Logic: We don't spin up a full browser instance for every request (which is slow and expensive). We attempt lightweight fetches first, escalating to full browser automation only when the site requires JS execution/hydration. - Token Optimization: Raw HTML is noisy and burns context window tokens. We process the DOM to strip non-content elements and return a markdown-friendly structure that is optimized for LLM consumption. - Infrastructure Stability: Scaling headless browsers is notoriously hard (zombie processes, memory leaks, crashing instances). We manage the fleet lifecycle and orchestration so you can run thousands of concurrent requests without maintaining the underlying grid. On Ethics: Since we are backed by Mozilla, we are strict about how this interacts with the open web. - We respect robots.txt rules. - We identify our User Agent. - We do not use requests/content to train models. - Data is ephemeral and discarded after the task. The linked post goes into more detail on the infrastructure and why we think browsing needs to be a distinct layer in the AI stack. This is obviously a very new space and we're all learning together. There are plenty of known unknowns (and likely even more unknown unknowns) when it comes to agentic browsing, so we’d genuinely appreciate your feedback, questions, and tips. Happy to answer questions about the stack, our architecture, or the challenges of building browser infrastructure.
More ai this month
the category →
I trained a 125M-parameter transformer to autocomplete piano performances in real time (~108 notes/sec on an iPhone 15). The idea is basically GitHub Copilot or Tabnine, except instead of prompting it with code, you prompt it by playing a few notes on a MIDI piano. The model then continues what you played, entirely on-device. The app is free if anyone wants to try it. Happy to answer questions about the model, training, Core ML, or the many things that didn't work.
AI · 16d ago · simedw.com
Astute▲585Automate your B2B brand going viral, with new media creators
AI · 18d ago · company-app.joinastute.com
Hey HN, Henry from Cactus here! We previously released Cactus Needle, a 14MB agentic LLM for tool call, device use, and structured extraction for phones, wearables, smart homes, small robots and microcontrollers. We got really great feedback here, and have now incorporated the suggestions to release Needle 2. The whole model is a single 14MB binary that runs a full session in 28MB of RAM; 45m parameters at 2bit compression. Needle hits 500 tokens/sec decode speed on a Raspberry Pi 5, sits between 400-1,500 tokens/sec on VR devices like Meta Quest 3S and Apple Vision Pro, and ranges…
AI · 26d ago · cactuscompute.com
Launched alongside, January 2026
the whole month →- IN
Hey HN! I wanted to share something I built over the last few weeks: isometric.nyc is a massive isometric pixel art map of NYC, built with nano banana and coding agents. I didn't write a single line of code. Of course no-code doesn't mean no-engineering. This project took a lot more manual labor than I'd hoped! I wrote a deep dive on the workflow and some thoughts about the future of AI coding and creativity: http://cannoneyed.com/projects/isometric-nyc
AI · Jan 2026 · cannoneyed.com
Automatic AI-powered code reviews the moment you open a PR
Dev tools · Jan 2026 · kilo.ai
