SuperAPI AI Directory
Partner with us中文AI newsFavorites
All toolsAI appsRill

Rill

Rill is a free browser for Mac that lets Claude Code and Codex agents work alongside the user while browsing the web. Users can give the agent a task or ask a question on any page with ⌘E, then keep browsing while the agent completes the task in a background tab. It directly uses the user's existing Claude Code or Codex subscription, with no extra payment required. The browser features automatic grouping of agent tabs, a daily task summary, a cross-project map, and supports voice input (Beta). Regarding privacy, history and passwords are stored locally, and AI features run locally under the user's own account. It is currently in Beta and only supports Apple silicon with macOS 14 or later.

Open website

About this tool

Rill is a free Mac browser whose core feature is bringing Claude Code and Codex agents directly into the browsing experience, letting AI work "beside" the user while they browse the web.

Purpose

Rill solves the problem of frequently switching apps when using AI coding agents. Instead of leaving the current page to open a separate tool, users issue instructions right where they find inspiration or a task. The browser knows the user's project context, and each request automatically lands in the correct project. Non-coding tasks (such as checking flights) can also be handed to the agent in the same way, with the search completed in a background tab.

Main capabilities

  • Send tasks or questions from any page: shortcut ⌘E, and the agent responds beside the page or in a background tab.
  • Tabs automatically grouped by AI: open tabs are automatically categorized based on what the user is doing, keeping the tab bar tidy.
  • Daily agent work summary: see what the agent accomplished today.
  • Agent work status panel: shows which agents are working and which models they are using.
  • Automatic project recognition: discovers projects from the user's Claude Code and Codex history, with no manual setup required.
  • Task completion notifications: the user does not need to check actively; the agent lets them know when it is done.
  • Web task execution: for example, for booking-type tasks it says "book it," and the agent executes up to the payment step and then stops.
  • History as a log: browsing history is presented as a record of "where you went, what you asked, and what was completed," supporting fuzzy page lookup by description.
  • Cross-project map: after a few weeks, users can see on a single map all the projects the agent has touched, with completed work marked as Done.
  • Voice input (Beta): hold Fn to speak an idea and the coding agent starts working; one-time setup is required in Settings › Talking.
  • Works as a normal browser when there is no agent: supports basic browsing features.
  • Page Q&A: ask questions about the current page, and answers open beside the page through the user's existing agent.
  • Tab comparison: select up to six tabs and generate a table citing the content of each page.
  • In-project agent reads web pages: browses in a separate tab, without using the user's logged-in state, and the user can take over at any time.
  • Basic browser features: passwords stored in Keychain, private windows, importing login information from Chrome, Arc, or Safari.

How to use

  1. Download Rill and drag it into the Applications filefolder.
  2. Sign in to Claude Code or Codex; Rill will guide you through it, and you can use one or both.
  3. Press ⌘E on any page, and Rill will automatically find your project.

Agent features require Claude Code 2.1.261 or later, or Codex CLI 0.144.4 or later. Voice features need to be configured in Settings › Talking. Tab sorting and daily summaries can be turned off in Settings.

Use cases

  • Hand off inspirations and tasks to a coding agent at any time while browsing the web.
  • Have the agent complete tasks such as searching, querying, and comparing page content without leaving the current page.
  • Manage multiple projects and want agent requests to be automatically routed to the correct project.
  • Quickly give instructions to a coding agent by voice.
  • Need to review the agent's daily work and track completed tasks by project.

Limitations

  • Currently in Beta and free.
  • Only supports Apple silicon Macs, requiring macOS 14 or later.
  • AI features depend on the user's existing Claude Code or Codex subscription; no new subscription is provided.
  • Voice input is in Beta and requires one-time setup.
  • New tasks are read-only by default until the user allows editing; in a git repository, each edit can be undone.
  • Web tasks will be asked to stop before executing steps such as payment, placing orders, sending, publishing, or deleting, unless the user explicitly instructs it to continue; the agent cannot enter passwords, card numbers, or one-time verification code fields.
  • The agent treats pages as untrusted text; in projects, the agent browses in a separate tab, does not use the user's logged-in session, and only clicks or types when the user allows it. Web tasks use the user's logged-in session but are subject to the above limitations. These measures reduce risk but cannot completely eliminate it.
  • History and passwords are stored locally on the user's Mac: history is kept as files for a duration chosen by the user; passwords are stored in the Mac Keychain and unlocked via Touch ID.

Privacy and security

AI features run Claude Code or Codex on the user's Mac under the user's own account. History and passwords do not leave the local machine. Tab sorting and daily summaries can be turned off. For the specific content sent by each feature, official detailed documentation links are provided.

Related tools

Browse category