PRODUCT REVIEW · 9 OCT 2026
Rill connects a Mac browser to the Claude Code or Codex CLI you already use. Here is what its beta promises, what remains untested, and when Tabbit may be a better fit.
Based on Rill’s public documentation and product demonstrations. We have not installed or benchmarked Rill.
THE SHORT ANSWER
Rill is compelling if your next step from a webpage is usually work in a local code project. Tabbit is worth trying if you want an AI browser for everyday web research and browser tasks, especially on Windows.
Made by Manifold AI, Rill is a free early beta for Apple silicon Macs running macOS 14 or later. Press ⌘E on a page to send its context to your existing Claude Code or Codex setup. Rill says it can find the right local project, let the agent work while you browse, and report back in the browser. Its other features include AI tab groups, a day-based history, page questions, and comparisons of up to six tabs.
Attach the current page or a selection and describe the task with ⌘E.
Rill chooses a local project or a web task, then runs your signed-in Claude Code or Codex CLI.
The agent runs in the background and asks for input or shows a result in the browser.
Rill looks through your Claude Code or Codex CLI project history to suggest where a task belongs, and asks if it is unsure. New project tasks start read-only until you allow edits; the interface shows tool steps and file diffs, with undo for changes in Git repositories.
For a request such as finding a flight, Rill can work in browser tabs instead of a code project. A web task can use your normal signed-in tab; browser tabs opened by a project agent have separate temporary cookies. You can stop a task with ⌥⌘.
These are examples shown by Rill, not tasks we reproduced.
A research paper
Ask an agent to apply its method to data in a local research project and make a chart.
A release note or table row
Send the selected update or row into the relevant code project for a test or comparison.
A city guide
Start a web task to look for flights while you continue reading other tabs.
Rill also tries to organize what you browse and help you find it later.
Rill groups open tabs by topic without replacing the normal tab bar. It can do this automatically or on demand, and automatic grouping can be turned off.
History is organized into daily summaries. Exact-word search runs locally; Describe search uses an agent to find pages by what you remember about them.
A project map groups projects found in your CLI history. A corner panel shows running agents, questions that need your input, and completed work.
Ask about the current page in a side panel, or select up to six tabs for a cited comparison. Talking Beta also lets you hold Fn to dictate a task after setup.
Running the CLI on your Mac does not mean every AI request stays there. Rill’s privacy page describes different data sent for each feature.
The comparison is about documented workflows and availability, not speed or task success rates. Both products can change during beta.
| Decision point | Rill browser | Tabbit browser |
|---|---|---|
| Best starting point | A page that should become a task in a local coding project | A page or set of pages to research, summarize, or act on in the browser |
| AI setup | Requires an installed, signed-in Claude Code or Codex CLI for AI features | AI chat and Agent mode are built into the browser |
| Platforms | Apple silicon Mac, macOS 14+ | macOS and Windows desktop |
| Web context | Current page, selection, and up to six tabs for comparison | Browser pages can be supplied as context for chat, research, and Agent tasks |
| Pricing at review date | Rill beta is free; existing agent-account usage still applies | Tabbit offers a free browser download |
Tabbit puts chat and Agent tasks beside the browser instead of making an existing coding CLI the starting point. These screenshots show the product interface and example workflows; they are not a head-to-head test against Rill.
The new tab has a prompt box, page references, tabs, and a conversation area in one window.
A Deep Research task can sit next to the page being read, with visible research steps.
Agent activity can be followed alongside a website such as this product-management page.
We did not test Rill’s task completion, resource use, extension support, or its security controls. Rill’s privacy page says AI features send different amounts of page or project information to the selected model provider. Its “stop before payment or sending” rule is an instruction to the agent, not a technical lock. Review the official privacy page before connecting sensitive work.
Choose Rill if you use Claude Code or Codex daily on a supported Mac and want to turn browsing into local project tasks. Try Tabbit if you want integrated web research and Agent workflows, or you need Windows. Test both with your own sites and tasks before moving your main browsing workflow.
Rill’s website calls its current beta free with no extra Rill subscription or API key. AI work uses your existing Claude Code or Codex account and its applicable limits.
The reviewed release lists Apple silicon Macs with macOS 14 or later. Rill does not list a Windows download. Tabbit has a Windows desktop version.
No. Rill says local history and passwords stay on the Mac, but AI features send relevant page and project content to the selected Claude Code or Codex provider. The amount depends on the feature.
Sources and review scope
Information checked 9 October 2026
Explore Tabbit’s integrated chat, research, and Agent workflows on macOS or Windows.
