---
type: "post"
id: "browseros-open-sources-neo-a-browser-for-coding-71629c"
slug: "browseros-open-sources-neo-a-browser-for-coding-71629c"
title: "BrowserOS open-sources 'neo', a browser for coding agents with per-agent tabs and session replays"
description: "BrowserOS has open-sourced neo, a browser environment intended for coding agents."
retrieval_nugget: "BrowserOS has open-sourced neo, a browser environment intended for coding agents. BrowserOS says agents can receive their own tabs, import Chrome logins, and save sessions as scrub-able replays. For teams operating browser automation, the proposition is a more inspectable shared runtime rather than another chat interface."
published_at: "2026-11-16"
updated_at: "2026-11-16"
record_date: "2026-11-16"
date_kind: "published_at"
topics: ["coding-agents","claude","claude-code"]
entities: ["BrowserOS","neo","Claude Code","Codex"]
related_patterns: ["harness-architecture-outlives-model-choice"]
novelty_label: "new_fact"
what_changed: "BrowserOS open-sourced neo, a coding-agent browser with per-agent tabs and session replays."
prior_refs: []
source_url: "https://github.com/browseros-ai/BrowserOS"
source_urls: ["https://github.com/browseros-ai/BrowserOS"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:browseros-open-sources-neo-a-browser-for-coding-71629c"
status: "published"
visuals: []
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
routes: {"html":"https://newruntime.com/posts/browseros-open-sources-neo-a-browser-for-coding-71629c/","markdown":"https://newruntime.com/posts/browseros-open-sources-neo-a-browser-for-coding-71629c.md","json":"https://newruntime.com/posts/browseros-open-sources-neo-a-browser-for-coding-71629c.json"}
markdown_url: "https://newruntime.com/posts/browseros-open-sources-neo-a-browser-for-coding-71629c.md"
---

# BrowserOS open-sources 'neo', a browser for coding agents with per-agent tabs and session replays

## Retrieval answer

BrowserOS has open-sourced neo, a browser environment intended for coding agents. BrowserOS says agents can receive their own tabs, import Chrome logins, and save sessions as scrub-able replays. For teams operating browser automation, the proposition is a more inspectable shared runtime rather than another chat interface.

## Summary
BrowserOS has open-sourced neo, a browser environment intended for coding agents. BrowserOS says agents can receive their own tabs, import Chrome logins, and save sessions as scrub-able replays. For teams operating browser automation, the proposition is a more inspectable shared runtime rather than another chat interface.

## What happened
BrowserOS’s GitHub repository presents BrowserOS as an open-source agentic browser and describes it as an alternative to ChatGPT Atlas, Perplexity Comet, and Dia. The project’s neo environment is aimed at coding agents that need to work on the web, not only at a person browsing manually. BrowserOS names Claude Code, Codex, and MCP-based agents as systems that can use the environment.

The release’s central operating idea is to give each agent its own tabs. Instead of placing an agent into the same browsing context a person is actively using, neo is described as assigning a dedicated tab set to that agent. That is a concrete change in how browser work can be organized: a human can retain a normal browser session while an agent handles a web task, and separate agent runs can be kept apart at the tab level.

This matters in workflows where browser state is part of the task. A research run, a support operation, a test of a web application, and an authenticated back-office flow can all depend on the current page, navigation history, cookies, and the sequence of actions taken. Dedicated tabs do not by themselves prove a security boundary, but they offer an explicit unit for separating work. For a team running multiple tasks, that can make it easier to identify which run produced a particular browser outcome.

BrowserOS also says Chrome logins can be imported in one click. The intended result is that an agent can work in services where a user is already authenticated instead of requiring a new sign-in routine for every task. That makes authenticated browser automation more immediately useful, particularly when a workflow crosses sites that an employee already uses. It also means the browser environment should be treated as a privileged operational surface: an imported session can carry the authority of the account behind it.

The other notable feature is session replay. BrowserOS says every session is saved as a replay that can be scrubbed through. A replay creates a possible review path after a run: an operator can follow the sequence of browser activity, locate where a task went wrong, and distinguish a bad instruction from an unexpected page state or an unreliable action. The repository positions neo as a shared, replayable browser environment for agents rather than simply a mechanism for issuing browser commands.

Together, per-agent tabs, imported logins, and replay point to a browser being treated as part of the agent harness. The model may decide what to do, but the runtime holds the visible state, the access path, and a record of what occurred. BrowserOS has made that environment available in an open-source repository, giving teams a project they can evaluate rather than only a product claim.

## What the archive adds
New Runtime’s earlier coverage on December 16, 2025 described Browser Use expanding from an open-source browser-control library into a broader mix of open-source and managed components. This is a new event: neo makes browser execution itself more explicit through per-agent tab allocation and replayable sessions.

## Trend signal
This supports [Harness architecture outlives model choice](https://newruntime.com/patterns/harness-architecture-outlives-model-choice/). BrowserOS’s proposition is not a new model; it is the browser layer around models and coding agents. Dedicated workspaces, authenticated state, and reviewable traces can remain relevant even as a team changes the model or agent interface that drives them.

It also gives practical shape to the idea that agent reliability is partly a systems problem. A browser task is not only a prompt and a completion. It is an execution environment with state, access, and an audit trail. The release does not establish that neo solves those problems in production, but it makes them first-class product surfaces.

## Why it matters
Browser-based agents often fail in ways that are hard to reconstruct. A final answer may say that a task was completed while the underlying run encountered an unexpected login screen, selected the wrong tab, changed page state, or stopped after an unsuccessful click. When the only artifact is a chat transcript, a developer has limited evidence for reviewing what actually happened in the browser.

A replayable session can improve that review loop. It gives a developer or operator an artifact to inspect after a run, which can support debugging, regression investigation, and approval of consequential actions. Per-agent tabs can complement that record by making concurrent browser work easier to attribute. Those are operational benefits rather than guarantees of correctness, but they address a recurring pain point in browser automation.

Imported Chrome logins make the trade-off sharper. Removing sign-in friction can unlock useful workflows, but it also brings existing account authority into an agent-operated environment. Teams should start with a limited account or a narrowly scoped test environment, identify which sites and actions the agent can reach, and review a replay before allowing unattended use. Treat browser access as an authorization decision, not as a convenience setting.

## Limits and uncertainty
BrowserOS’s release is a product announcement and repository description, not independent validation of reliability, security, or performance in production environments. The operational behavior of replay, the fidelity of recorded sessions, retention and export practices, and the scope of browser-state isolation all require direct evaluation by adopters.

Teams also need to determine how imported Chrome sessions behave in their own identity and security setup. An authenticated browser can expose sensitive information or enable meaningful account actions. The release establishes a project and its stated capabilities; it does not remove the need for threat modeling, access controls, test accounts, and human review around high-impact browser tasks.

## Skill to build now
Create a small browser-agent evaluation harness this week. Choose one low-risk task, run it with a dedicated test account, and define a pass condition outside the agent’s own narration: a changed test record, a downloaded artifact, or a verified page state. Then review the session replay against that condition.

Use a separate browser context for each trial and log the account, task goal, expected end state, and whether a human approved any irreversible action. This makes replay useful as operational evidence rather than a visual novelty. If the replay cannot explain a failure clearly, the workflow is not ready for sensitive automation.

## Reader takeaway
Neo’s notable proposition is not simply that an agent can browse. BrowserOS is packaging separate tab contexts, existing authenticated access, and replayable runs into one open-source browser environment. That is worth testing for browser-agent workflows, provided teams treat imported sessions as privileged credentials and make review part of the workflow.

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- topic hub: [Coding Agents](https://newruntime.com/topics/coding-agents.md): Explore the coding-agents topic hub.
- topic hub: [Claude](https://newruntime.com/topics/claude.md): Explore the claude topic hub.
- topic hub: [Claude Code](https://newruntime.com/topics/claude-code.md): Explore the claude-code topic hub.
- collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis.
- collection: [Sources](https://newruntime.com/sources/): See the original sources this archive cites.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)
