Talona

Browser automation using AI agents.

[Kacper Gadomski @k4cperg]

We killed our product.

You might need to do it too.

We built @TalonaAI as AI browser infrastructure. One agent, any website, any task. The bet was width: scraping, QA, RPA, research, sales.

Then we watched the runs. Almost nobody used it as a browser for agents.

They used it to find leads and run outreach. LinkedIn. X. A buying signal. A message. A follow-up. A meeting. Same loop, every time.

Most teams treat that as noise and keep shipping the original vision. Width feels like optionality. It is usually a delay. You stay useful to everyone and essential to no one.

The data was already a niche. We were last to accept it.

So we killed the wide product. @TalonaAI is now an AI GTM agent. It finds people who need you now, starts the conversation, and puts the meeting on your calendar.

If your last 50 sessions collapse into one job, that job is the company.

If that job is finding customers, visit https://t.co/mH1LrkEsla

[Kacper Gadomski @k4cperg]

our stats 2 weeks after launching @TalonaAI beta:

> 40 daily users
> 1.2k page visits
> 21m avg session duration
> 20k views across social media

pretty cool, taking into account we payed 0$ on advertising.

go check it out at https://t.co/2iJEVcJWBB

[Kacper Gadomski @k4cperg]

Introducing Gemini 3.8 Flash for Web Automation.

Gemini 3.8 is the most efficient model for automating your browser tasks, along with being dirt cheap, now powering Talona's browser agents.

Just like Fable, Gemini is also available for FREE for the next 48 hours on https://t.co/y7ypsD3UJT

[Ignacy Ruszkowski @ignacyrusz]

grab it while you can!

go to https://t.co/2U1gFMSZ1s and sign up today to get free access to Fable 5.1, offer ends in 37 hours!

[Kacper Gadomski @k4cperg]

Introducing Fable 5.1 for Web Automation.

Fable is both the most advanced and cost efficient model to automate your browser with, now powering @TalonaAI's browser agents.

Available for FREE for the next 48 hours on https://t.co/rOstTSaPQT

[Kacper Gadomski @k4cperg]

We had a problem. Our browser agents were unusually slow.

We thought the browser automation harness was too heavy, so we trimmed startup work and optimized the obvious paths.

Startup: 1.73s → 0.97s (-44%)
To-first-action: 4.74s → 4.30s (-9%)

It helped, but not enough.

So we instrumented the entire path:
agent → browser harness → CDP → Chrome

Turns out the biggest bottleneck was our harness written in Python.

So we re-wrote it in Rust.

The Rust rewrite removed Python startup, stdin/stdout IPC, BrowserSession initialization, and extra browser-control handoffs.

We also retained earlier optimizations like attaching to the existing Chrome instance and using one startup tab.

Across five production VM runs:

Readiness: 1.62s → 1.14s
To-first action: 4.05s → 3.20s
Warm observe: ~5–7ms
Warm action: ~3–5ms

29% faster readiness.
21% faster end-to-end.

[Ignacy Ruszkowski @ignacyrusz]

As a founder, I spend huge amounts of time reaching out to people. I have to look them up on different websites, check if they are actually relevant, contact them afterwards. There's a lot of time wasted in this process, and that's where https://t.co/VZCHbN2UtH comes in!

With out current capabilities you can automate this research part entirely by delegating it to Talona. I found this improved my workflows significantly and freed my time for more important things. Look at the demo to see it in action below!

At the same time, thanks to the feedback of our first users, we made significant improvements to our website design. Check out if Talona can help you automate your workflows at https://t.co/VZCHbN2UtH.

[Kacper Gadomski @k4cperg]

We gave @TalonaAI agents nothing but a description of the data we wanted. No URLs.

They found the freight exchanges, got through the logins, and came back with the fields.

We'd spent months finding those sources by hand. The agents did it in 10 minutes.

209,342 truck postings, 1,370 corridors, now open.

Link in comments.

[Ignacy Ruszkowski @ignacyrusz]

getting backed by @reaktorx was not something i had in mind 2 months ago, yet here i am. when life throws opportunities your way, you start thinking. i am glad that i found myself surrounded with people who believed in me, and was able to start executing.

i've always embraced risk-taking and stepping outside of my comfort zone. this step takes me further than i've ever been before.

together with @k4cperg we aim to change how agents use the web, you can check it out NOW at https://t.co/VZCHbN3sjf. if you are in San Francisco in October and would like to talk, shoot me a dm.

thank you @k4cperg @luca_stirbat

[Kacper Gadomski @k4cperg]

we just got into @Reaktor_X

we’re building an AI infrastructure company @TalonaAI

our goal is to make ai agents interact with the web blazing fast and dirt cheap

me and @ignacyrusz will be in sf through october, dm’s open