Show HN: Browser AI agent platform designed for reliability | Hacker News Reader

Show HN: Browser AI agent platform designed for reliability

github.com

76 pointsogandreakiro40 comments

We’re very excited to share something we’ve been building. Notte https://www.notte.cc/ is a full-stack browser agent platform built to reliably automate a wide range of workflows.

Browser agents aren’t new, but what is still hard is covering real-world flows reliably. The inspiration for Notte was to make a full-featured platform that bridges the agent reliability gap. We’ve packaged everything via a singe API for ease of use:

- Site Interactions - Observe website states, scrape data and execute actions

- Structured Output - Get data in your exact format with Pydantic models

- Stealth browser sessions - built-in CAPTCHA solving, proxies, and anti-detection

- Hybrid workflows - Combine scripting and AI agents to reduce costs and improve reliability

- Secrets vaults - Credential management to store emails, passwords, MFA tokens, SSO, etc.

- Digital personas - Digital identities with unique emails, phones for account creation workflows

With these tools, Notte allows you to automate difficult tasks like account creation, form filling, work on authenticated dashboards. Close compatibility with Playwright allows you to cut LLM costs and improve execution speed by mixing web automation primitives and include agents only for specific parts that require reasoning and adaptability.

Here’s a short YouTube demo: https://www.youtube.com/watch?v=b1CzmfpdzaQ

If any of this sounds interesting, you can run your first agent following our quickstart on GitHub https://github.com/nottelabs/notte. Or play around with our free plan through our Notte Console: https://console.notte.cc/

We’d love to hear if there’s anything else required before you’d try or trust it on your own workflows :)

Open articleView on HN

Do you support robots.txt so administrators can opt-out of your service?

Not yet - this is a TODO and we'll try to do it asap!

Avoiding captchas and disrespecting robots.txt. How does it feel to advertise your spam service? Are you proud?

The pricing page was higher in priority

We added the pricing breakdown here: https://www.notte.cc/#pricing

Hey- we built Notte for legitimate workflows like testing and approved automation; not abuse. Like other session provider tools (Browserbase, Hyperbrowser, Anchor, etc.), we encourage respecting site terms and robots.txt, and we’re adding built-in support for it soon to enforce this policy inside Notte.

You should also support Web Bot Auth and make it required to have a customer-specific abuse contact in the user-agent header so that people can send invoices to those customers for when your customers cause downtime due to excessive scraping.

Yes, that’s a very good point! We’ll look into this for SDK usage. What do you recommend for the public repo where it's harder to enforce runtime headers?

No compliance? No support.

The pricing page mentions how many credits you get but not what a credit does or gets you. Could you elaborate on that?

Oh thanks for pointing this out. Pricing is explained on our landing https://notte.cc.

It’s 500 free credits. Pro plan $79/mo for 10K credits/mo. You can do additional top up 1K credits for $10 when in Pro plan.

If you need more it’s a custom enterprise deal - we work a lot with volume discounts.

To reiterate, "what a credit does or gets you" is not explained by your pricing page.

I'll operate on the assumption that it takes roughly 500 credits to scrape a single page for now.

Oh - missed that too. We had this somewhere in the docs and got lost.

- 1 scrape URL : 1 credit ~ $1c - 1 agent step : 2 credits ~ $2c - 1 min browse-time: 1 credit ~ $1c

Meaning if you run an agent for a 10-steps task for a minute it'll cost ~ $21c

It's simple! For $50 a month we give you a thousand smeckles which you can use to crank the floba. A very good deal.

Just explained below what you can do with the 10K smeckles - that's 10K URL scrapes or 5K agent steps or 10K minutes of browsing :)

Update- https://www.notte.cc/#pricing

But with AI it never tells you that. I pay $xxxx for the Claude API, half of the time it breaks it's own code and just does things explicitly told it NOT to do; I think I want to be refunded for the time wasting and frustration, but that's not happening. So you pay credits and you get 'stuff' that 'might work'; that's the reality. I'm gonna say this is basically not even allowed for consumers in the EU, but I don't know that for sure.

Pricing credits - dollars is explained below. +We were thinking about not charging credits whenever one of our agents fails because of a random error on our side (not too frequent, but still). This makes it more reasonable to start exploring and building with us.

Makes sense, thanks.

Just want to congratulate all involved for the obvious amount of care that has gone into thinking this through ...

I really hope it succeeds through the roof.-

Thanks - that’s nice to hear! :)

I've got to say - and this in a way makes sense: The model seems to be very "URL-oriented" ...

... as in (at least in the demo, the search one) if you ask it to visit a site by name (ie. "Go go Altavista and ...," ...

... it will complain "no URL" has been provided.-

It probably has to do with the "special instructions" given to the AI for this one program/demo ...

... still.-

Yeah - the search demo (https://search.notte.cc/) is just an attempt to showcase what one can do with a scraping agent embedded in a served LLM with an MCP server. We have a sys prompt pushing the LLM to request an URL to make sure it's scraping what you want to scrape - but I've played with it a bit and it can also come up with it's own URLs to be scraped.

Looks good, excited to try this together with the CDP feature. If it works well we can add it to our (TestingBot) integration/documentation.

Hey! Sure happy to create a shared integration with TestinBot - shoot me an email so we can discuss andrea@notte.cc :)

for the hybrid workflows, curious how do you decide which parts need AI reasoning vs can be hardcoded? is it adaptive or manual config?

right now it's hardcode - you write a web automation script where at some parts you have a very specialized agent to take the very few steps in your workflow that require reasoning/adaptability. Future: we're trying to make this process automatic.

Credit based pricing is broken. No way u produce shiy then user get out of credit to get nothing done.

There needs to be a way to purchase "X" more credits then it works IMO

There is!

There's a way to top-up 1K credits for $10 in our Pro plan. But if someone needs more then the 10K/mo + manual topups - I'd suggest moving to a custom deal. We can unlock any credits you need with custom rate limits and volume discounts.

You missed the point. What i mean, is your pricing scheme is a scam because u never know if your bot response would solve customer issue. Stop this madness, ok >?

we’re working on fail-credit refunds so charges only apply to successful runs

just tried to use it to extract data from hyatt.com

completely failed

another hype but actually doesn't work browser agent.

hey - thanks for trying out. not sure what you're trying to extract from hyatt.com but just naively tested the most basic agent (https://github.com/nottelabs/notte?tab=readme-ov-file#run-in...) with simple prompt task="scrape data from hyatt.com landing page" and it worked perfectly:

answer=Successfully navigated to hyatt.com and closed the cookie consent. The page content includes navigation links (Explore, Offers, Meetings & Events, Loyalty Program, Language, Sign In or Join), a search form with inputs for City/Address/Landmark/Airport/Hotel, Check-in and Check-out dates, and options for rooms/guests and special rates. Below the search form, there are sections for 'Earn bonus points at new hotels', 'Limited-Time Offer: Earn 2 Free Nights + 25,000 Bonus Points', 'More Places to Discover' with specific hotel examples (Park Hyatt Auckland, Hyatt Regency London - The Churchill, 7Pines Resort Ibiza, Hyatt Centric Murano Venice), 'Join World of Hyatt' with benefits, and various promotional sections like 'Incredible offers, incredible places', 'Plan a summer of more', 'Outdoor adventure awaits', and 'Wellbeing your way'. The page also lists different Hyatt brands categorized into Luxury, Lifestyle, Inclusive, Classics, and Essentials. At the bottom, there are links for Reservations, Customer Service, Corporate Sites, Explore More, and Connect with Hyatt (social media links), along with legal information (Privacy Policy, Terms & Conditions, Cookie Center, Security & Safety, Modern Slavery and Human Trafficking, Do Not Sell or Share My Personal Information)

How well does it handle captchas?

There’s no magic here unfortunately. We’re working hard to make anti-detection as good as possible with proxies to avoid captchas in first place. When they get there - we can solve ~60% of providers right now (incl reCAPTCHA, Cloudflare, and main ones) and some others are still work in progress.

does your scraping endpoint also provide stealth mode?

Yes! We just added stealth support for our scraping endpoint. The docs should be updated soon but you can set proxies now

no typescript support?

Reply on news.ycombinator.com