Every developer knows the friction: you find a release note, a pricing page, or a slick UI component somewhere on the web, and the interesting part — getting your coding agent to actually do something with it — costs another ten minutes of copying context into a terminal. Rill Browser is an AI browser for coding agents built to erase that handoff. Press ⌘E on any page and Rill turns what you are looking at into a task for your existing Claude Code or OpenAI Codex agent, picks the right project from your agent history, and passes the page context along. The agent works away in a background tab while you keep browsing. It is free while in beta, rides on the plan you already pay for, and is currently macOS-only on Apple silicon. Our verdict, after reviewing the launch and the early independent coverage: this is the most practical take on the agentic browser we have seen — but platform limits and beta-stage rough edges mean most teams should treat it as an experiment for now, not infrastructure.
What Rill Browser does: an AI browser for coding agents, not a chatbot sidebar
The distinction matters. Most so-called AI browsers bolt a chatbot onto the sidebar: you ask questions about the page, it answers. Rill’s premise is different — the browser does not do the work, your agent does. Rill is the bridge between the page you are reading and the codebase your agent can change.
The core loop is deliberately small. Browse normally, then press ⌘E on anything worth acting on. Rill packages the page into a task, suggests the right project from your Claude Code or Codex history, and sends it off. The agent works in background tabs while you keep browsing. Nothing about your browsing flow changes until you decide to delegate something.
Around that loop, Rill does the tab hygiene every power user secretly wants: it sorts open tabs into themes, turns your history into searchable “days”, and pulls every project your agents are working on into one view. Useful, but secondary — the task handoff is the product.
The idea has landed. Rill launched on Product Hunt on 6 October 2026 and picked up 500+ upvotes in its launch week, according to early coverage — a strong signal for a Mac-only developer tool that the workflow resonates with working developers, not just demo-watchers.
How the ⌘E workflow actually works
Here is the concrete loop, as documented in the launch materials:
- Browse normally in the Rill macOS app. It behaves like a regular browser until you invoke it.
- Hit ⌘E on anything worth acting on — a dependency’s release notes, a UI pattern you want to copy, a competitor’s pricing page, a forum thread about an error you are seeing.
- Rill identifies the right project from your Claude Code or Codex history. Practical note: use your agent on real projects first, or the project picker has nothing to learn from.
- The page context goes to the agent as a task, and it runs in background tabs while you keep browsing.
- Sensible defaults are in place: new tasks start read-only until you explicitly allow edits, and web tasks stop before paying, posting, sending, or deleting anything.
Three workflows where this genuinely helps
1. Dependency upgrade triage
Release notes are the perfect input for this tool. Point the agent at the changelog and ask it to compare against your project’s package manifest: which upgrades are safe, which are breaking, and what the migration effort looks like. The page gives the agent the source of truth; your project history gives it the target. This is the workflow we would reach for first.
2. Competitor and design teardowns
Reading a competitor’s product page and thinking “rebuild this flow in our stack” becomes a queued task with the reference page attached — instead of a ticket with a vague link and a screenshot that goes stale. Design and product teams can collect references all week and hand the agent a batch.
3. Error investigation with live context
Instead of pasting an error message and a documentation page into the terminal as two separate steps, the agent gets the page you were already reading, with your project as the implicit context. It is a small saving per incident that compounds fast on a busy team.
Pricing: free browser, metered agent
Rill itself is free during the beta period. The agentic features run on your existing Claude Code or Codex plan — you do not buy a new subscription for the browser’s AI capabilities. That is genuinely attractive: most AI browsers want you on their own seat licence.
Here is the honest version of “free”, though: your agent usage is still metered by Anthropic or OpenAI under your existing terms. If you are on a Claude Code subscription or paying per token through Codex, none of that changes. Evaluate Rill’s cost as browser (nothing) plus your unchanged agent bill — and keep an eye on usage if you are on metered plans, because background tasks are very easy to fire and forget.
Who it’s for — and who should skip it
It is for you if you already live in Claude Code or Codex daily. The browser only adds value on top of an agent workflow you already trust. It also suits agency and product teams where research constantly turns into implementation — the research-to-task shortcut is a genuine time-saver — and solo builders doing heavy competitor scanning, dependency evaluation, and reference collecting.
Skip it if you are on Windows, Linux, or an Intel Mac — officially unsupported, full stop. Skip it if you do not use Claude Code or Codex, because without a signed-in agent the core feature is inert. And skip it if your projects are small enough that the agent already sees everything; the context handoff solves a problem you do not have.
The honest limitations
- Narrow platform support. macOS 14 or later on Apple silicon only, plus Claude Code 2.1.261+ or Codex CLI 0.144.4+ installed and signed in. That is a small slice of the developer population, and mixed-platform teams cannot standardise on it.
- It is a 0.1.x beta. Treat it like one: try it on a side project first, not your main repo, and certainly not a client’s codebase until you have measured its behaviour.
- Project matching is inference, not magic. If Rill picks the wrong project from your history, an agent with file access starts reading the wrong codebase. Verify the target project before granting any edit permission.
- Prompt injection is the real risk. A browser feeds untrusted web content to an agent that can touch your files. To Rill’s credit, its own site says its safeguards “lower the risk” rather than removing it — refreshingly honest for a launch page. Keep the read-only default, and treat pay, post, send, and delete as manual-only, always.
- The privacy surface widens. Page context, source code, prompts, and agent output now flow through Rill plus your agent provider. Before pointing it at client work, check your contracts and your organisation’s data policies — “it was visible in my browser” is not a data-handling policy.
How Rill compares to the other AI browsers
Dia (from The Browser Company) and Perplexity’s Comet are full browsers with built-in AI assistants: you talk to the browser, and the browser browses for you. Rill inverts the relationship — it stays out of the browsing and hands structured work to the agent that already owns your codebase. If you have already invested in a Claude Code or Codex workflow, Rill asks less of you than adopting a whole new assistant.
Copilot-style sidebars and in-browser assistants are good at answering questions about a page, but they do not create executable, project-scoped tasks. That is the gap Rill fills. And tools that let agents drive browsers (such as Vercel’s agent-browser CLI) go the opposite direction — the agent operates the browser. Rill keeps the human browsing and the agent coding, which matches how most teams actually want the division of labour.
The honest positioning: Rill is the only agentic browser built specifically around the Claude Code and Codex terminal workflow that serious developers already use, rather than around a new assistant you have to adopt on top.
From our experience: how we would actually deploy this at an agency
After nine-plus years and 300+ projects, we are suspicious of any tool that claims to remove a step — usually it just moves the work somewhere else. Rill genuinely removes the copy-paste handoff, but the work moves to review: queueing tasks is fast, verifying them is the job. Budget your time accordingly.
The agency test we would run before letting it near client repositories: one week on an internal project, tracking two numbers — how often it picks the wrong project, and how much context we have to correct per task. If both stay low, it graduates to supervised client work. If not, it stays a research toy. No tool gets production access on launch-day enthusiasm.
Our standing rule for any agent tooling also applies here: read-only defaults stay on, destructive actions stay manual, and the agent never touches a repository the human has not opened that week. Rill’s defaults align with this, which is why it is worth the experiment.
Where it would not help us: deep refactors that need architectural judgement, or anything where the page context is ambiguous. The agent only knows what the page plus your history gave it — unclear context in, confident nonsense out. Review the task brief before it runs, the way you would review a junior developer’s brief.
Frequently asked questions
What is Rill Browser?
Rill Browser is a Mac browser built around Claude Code and OpenAI Codex. Pressing ⌘E turns the page you are viewing into a coding task for your agent, which runs in background tabs while you keep browsing. It also organises tabs into themes and makes your history searchable.
How much does Rill Browser cost?
The browser is free during beta, and its agentic features run on your existing Claude Code or Codex plan — no new subscription is required. Your agent provider’s usage terms still apply, so metered plans can still rack up charges from background tasks.
Does Rill Browser work on Windows or Linux?
No. Rill currently requires macOS 14 or later on Apple silicon, plus Claude Code 2.1.261+ or Codex CLI 0.144.4+ installed and signed in. Windows, Linux, and Intel Mac users are officially unsupported.
Which coding agents does Rill work with?
Claude Code and OpenAI Codex — the ones you already use, signed in locally. Rill finds candidate projects from your agent history, so it needs that history to exist before the project picker is useful.
Is it safe to hand web pages to my coding agent?
Mostly, with guardrails. New tasks start read-only until you allow edits, and web tasks stop before paying, posting, sending, or deleting. But web content is untrusted input to an agent with file access, and Rill’s own site says its prompt-injection safeguards lower the risk rather than removing it. Keep destructive actions manual and start on non-critical projects.
Rill Browser is a sharp, narrowly scoped tool: an AI browser for coding agents that does one workflow — page to agent task — better than anyone else has, and charges nothing for it during beta. The Mac-only, beta-stage, agent-dependent reality makes it an experiment, not infrastructure, for most teams. If you already live in Claude Code or Codex on a Mac, it is worth the download and the one-week test. And if you are evaluating AI tooling for your team more broadly, browse our AI tools directory, read about our AI integration services, or talk to us — we have spent years separating the tools that survive contact with real client work from the ones that do not.