Nimbus - Agentic Browser with Claude Code UX
by•
Nimbus is an agentic browser inspired by Claude Code UX. Tell it what you want in plain English.
It handles the mechanical work: clicks, forms, file pickers, multi-tab coordination and pauses to ask only when there's something only you can decide. Inspired by the way Claude Code feels for coding; built for the rest of the web.
Built from the ground up, not an extension. Free for the first 500 founding users forever.
Replies
Nimbus
Hello PH, I'm Anil.
Most of my day is split between a terminal and a browser. And the browser part started becoming increasingly annoying, while on the terminal so many complex things happen with such a simple UX.
When I'm doing something complex on the web, I just want to think about the problem at hand. Instead I'm fighting the mechanical stuff navigating across tabs, finding the right files, copy-pasting between sites, dealing with how every site implements basic things differently.
The thing I hate most: I download a file in one tab, then have to upload that same file in another tab. Every browser implements the file picker differently. And nothing annoys me more than triggering the native file picker to hunt for a file I literally just downloaded.
So I built Nimbus. The agent handles all the mechanical work, and gives you the freedom to think about the problem at hand.
Built from the ground up, not an extension.
First 500 users are free forever: Download for macOS from here https://usenimbus.app
Happy to answer any questions.
HasData
@2abstract4me claude code energy is the right north star, love the bet. is the underlying model literally claude or are you going model-agnostic, and if claude, are you using their browser tool support natively or your own scaffolding on top?
Nimbus
@sergeynagorny it is model agnostic. currently it’s byok. supports all three major providers. claude, openai, gemini. and also any openai compatible llm.
Nimbus
Listing out the features here:
Agent uses oracle to plan complex tasks on the web, take its help when stuck, and also to create skills based on existing sessions.
Sessions. Each task is its own session with its own tab(s) and history. Switch between them and let the tasks run in the background.
Ask-user tool. When the agent hits a judgment call (cost confirmation, ambiguous field, a captcha), it pauses and asks you in the chat, not in a page overlay. You answer, it resumes. I'm really obsessed with the ask user tool on Claude Code, and obviously I implemented it here also.
Bring your own key. Gemini, OpenAI, Anthropic, or any OpenAI-compatible endpoint. No server of mine in the loop.
Skills. Teach it or let it figure out a reusable flow and save it as skill to reuse it.
Auth handoff. When a login popup opens, the agent blocks, you complete the auth, the agent picks back up. I purposefully didn't automate things like auth/captcha, as the expectation of the current websites' implementations isn't automation.
Everything local. Traces of every run go to ~/.nimbus/traces/. No telemetry, YET. Nothing reaches my servers, you just contact the LLM providers directly. You can use traces.usenimbus.app to look at the actual llm calls, tool calls, and the agent loop from past sessions.
Merging an agentic browser with a Claude style UX is a smart move. How does Nimbus handle multi-tab reasoning when a task requires data from several different sites?
Nimbus
@rivra_dev You are correct, a single complex task, usually involves multiple tabs, and the agent will open and close tabs per session/task, and the agent does use all the opened tabs as "context" to finish the task at hand, like an actual user might do.
HasData
Nimbus
@nikita_naumov You are so kind. Thank you so much for the support.
the download-in-one-tab upload-in-another problem is so painfully real , I deal with this constantly when scraping data for college assignments and it's one of those things you don't realise is stealing your time until someone points it out. built from scratch rather than an extension is the right call too, extensions always feel like they're fighting the browser rather than controlling it. just one doubt when the agent pauses to ask me something mid-task, does it remember the full context of what it was doing when I reply, or does longer tasks risk losing the thread?
Nimbus
@shwet_gaur it doesnt lose context if u stopped it
midway. it knows how to resume with previous context. .
@2abstract4me good to know it holds state properly. going to try this for my next data scraping assignment honestly
The amount of time spent just being a human bridge moving files and data back and forth between different browser tabs is ridiculous. Shifting all that mechanical grinding to a text bar instead of doing a thousand clicks a day sounds like a massive quality-of-life upgrade for daily workflows.
Nimbus
@mithra_xavier yes its not just mechanical grind. for example when im trying to do something complex. some of the dashboards like ad campaigns, tax filing and etc are whole professions compressed into forms and buttons. if u see them u dont know what the right thing to do is, unless u r an expert urself. i found myself moving to and fro from claude and the actual website to do some things. wholly its a way for me to reduce friction, and have the freedom to think clealry without doing much mechanical work.
The download-then-upload loop is one of those frictions that's so normalized nobody talks about it but everyone hates it. 'Built from the ground up, not an extension' is doing a lot of work in that sentence — curious what made you go that route vs the extension approach technically?
Nimbus
@marco_pott i decided to do everything from the ground up because i see the url bar and a seperate chat bar as wasting space. this is why i said inspired by claude code ux, a single place for u say what u want to do and the agent does everything else. even though this project was technically challenging to implement, its mainly a ux experiment for me. getting the agentic loop right with multiple providers, figuring out the right tools to make available, the app itself are hard work. but i mainly want a test bed that i can control and know inside out, so that i can do more experiments regarding agent native web. and play some small role in shaping where its going. and this is just a starting point, and ao many exciting plans for Nimbus.