No Terminal Needed

Agent chat

Give a task in Chrome’s side panel and let a local coding agent work with your logged-in browser. The extension uses your installed CLI and its configured model service.

The agent compares plans on the page and replies with a table

Install chrome-use, register the native host with chrome-use extension install, and install the Chrome extension. Install and sign in to at least one supported agent CLI, then open the side panel.

First run from the Chrome Web Store

Added the extension before the CLI? Open the side panel anyway. It shows a setup guide instead of an error: on a Mac, Download for Mac fetches a signed, notarized disk image; open it and double-click ChromeUse, which installs the CLI for your user and registers the native host without a terminal or a password. On Windows and Linux the guide shows the install command with a copy button. The panel retries every few seconds and continues by itself once the CLI answers. Next it lists Codex, Claude Code, OpenCode and Pi: installed ones with the one sign-in command to run in a terminal, missing ones with a link. Check sign-in marks each installed agent Ready or Signed out (it starts the agent CLIs, so it waits for your click). It looks again when you return to the panel; Hide dismisses it for good. The guide runs only in extension pages and adds nothing to web pages. Details: Installation.

Start a conversation

Click the extension icon to open the side panel; Settings opens connection options. On macOS, Linux or Windows, choose Codex, Claude Code, OpenCode or Pi with your installed CLI's login. The host reports which agents it supports; older hosts may need an upgrade. Windows also supports skills, images, conversation options, schedules and configuration apply/restore. Stop, delete and option changes confirm the whole agent process tree exits. Private state uses a protected user/SYSTEM ACL; no administrator access is required. Codex uses app-server, Claude Code uses stream-json, OpenCode uses ACP, and Pi uses JSONL RPC.

Pages and models

Each conversation has its own browser session. The page tag above the composer attaches the current webpage to your next message; click Γ— to omit it. Model choices come from the selected CLI's catalog. Default model uses its configured model; Custom model ID in the οΌ‹ menu overrides the choice. Reload models from CLI refreshes the catalog. A listed model or reported login does not prove execution access.

One panel per tab

The toolbar button opens a side panel for the tab you click it on. Each tab has its own panel and its own conversation: switch to another tab and the panel shows that tab's conversation, or closes if you never opened it there; switch back and the earlier conversation is still there. A panel opened on a new tab starts a new conversation. Picking a conversation that another tab shows moves it to this tab. Closing a tab keeps its conversation in the history.

Where the agent works

A conversation in a tab's panel works in that tab by default: its first message lends the tab to the conversation, and the agent reads and acts on your page with your logins instead of opening its own window. Agent works in above the composer switches one conversation between This tab and Its own window; the default is in Settings β†’ Side-panel agent. A lent tab:

Open in a tab (the arrow button in the panel header) opens the current conversation as a full-page tab with a wider, centred column. The panel and the full-page tab can show the same conversation at once: both update live, and a message is sent only from the view you typed it in. A full-page tab has no page beside it, so its page tag names the last web tab you used in that window (pick another from the list), and nothing is lent by itself: click Let the agent use this tab to lend the page in the tag.

Lends last until Chrome restarts. Per-tab panels and working in the panel's tab need CLI 1.5.223 and extension 0.5.63 or later.

Tool access and approvals

The permission switch applies to one conversation. New chats start with Full access unless you saved another choice.

Clicking a task notification activates the conversation tab and focuses the window containing that tab, even when another Chrome window was last active or the current window is not yet known. A bound conversation tab takes precedence over the last-window fallback.

Queue, stop and resume

Replies stream into history, with tool rows that expand for details. Tool output does not prove the browser task succeeded. Messages sent during a response queue up; failure or interruption pauses the queue until Resume queue. After a failed or interrupted turn, a new message starts immediately in the same conversation and model; older queued messages stay paused until you resume them explicitly. Stop & send or Ctrl/Cmd+Enter waits for confirmed interruption or process exit before sending ahead of the ordinary queue. Reconnect reads history without resending.

Panel replies preserve fenced code literally. To show an example containing triple backticks, wrap it in a longer fence. A closing fence must use the same character, be at least as long as the opening fence, and have only whitespace after it.

History and deletion

Stop the response before confirming deletion from the header. Deletion removes panel history and the conversation's browser session, plus Codex rollout files, Claude session files/directories, OpenCode history through session/delete, or Pi session files inside the conversation folder. Codex's SQLite cache is untouched. Cleanup failures or unconfirmed tab closure are reported.

Each conversation row offers Initial page, Final page, and Delete conversation. Initial page opens the first recorded HTTP(S) URL; Final page opens the most recently recorded URL, including while a task is running. A matching recorded tab is focused, otherwise the saved URL opens in a new tab. Opening the initial page preserves the final address. Older conversations without an initial address report it as unavailable. The title switches chats; opening a page does not send messages, rerun tasks or restore unsaved input. Addresses stay in local extension storage. Deletion requires confirmation and reports cleanup failures; stop a running reply before deleting.

If Chrome replaces a saved Initial page tab, the conversation follows its replacement instead of opening a duplicate. The saved initial and final addresses remain separate.

Continue a terminal session (/resume)

In the /resume picker, changing the agent or search clears the previous selection immediately. Enter only selects a currently listed session; it does nothing while results are loading, unavailable or empty. Changing the agent/search or closing the picker also dismisses old confirmations and prevents earlier resume responses from opening the old conversation. Requests already sent to the host may still finish; fresh results remain selectable while they are pending.

Type /resume (optionally with search words) in the composer, or choose Resume a session… in the conversation menu, to continue a CLI session you started in a terminal. The picker lists that agent's recent sessions on this computer grouped by project folder: projects with the latest activity first, the current site's remembered project at the top, and sessions without a folder under Other. Each project shows its three latest sessions, with Show more for the rest; click a project to fold it, and hover it for the full path. Search matches titles, project names and paths and keeps the grouping. Use ↑↓ to pick, Enter to continue, Esc to close. Last activity comes from the times recorded in the session, not the file's modification time (Codex Desktop rewrites files when it imports other agents' sessions); a session is marked active elsewhere only when it wrote a record moments ago or has an unfinished turn. Codex subagent threads are not listed. sessions the panel started itself are left out, and a session already continued in the panel opens that conversation. Choosing one creates a conversation bound to the session. Nothing is sent until you write the next message, and the agent then answers with the earlier context.

A session written in the last minute, or held by a running Claude Code process, is marked Active elsewhere. Continuing it asks for confirmation first, because two programs writing one session can corrupt its history; after you continue a session here, stop using it in the terminal. The session's earlier messages appear in a collapsed Earlier in this session block. They are imported redacted, like every other history record, and capped at the last 40 messages; anything older or unreadable is marked Earlier history not recorded. The session's folder becomes the conversation's project when it passes the same checks as a chosen folder; if it is a system or home folder, Claude Code and Codex continue in the private folder and the panel says so, while OpenCode, which only continues a session in its own folder, refuses. The CLI's own session files are only read, and deleting such a conversation keeps the CLI session. Requires CLI 1.5.220.

Project folder and site ↔ project

By default a conversation's agent works in a private folder of its own. To work in a project instead, choose a folder in the folder chip next to the agent (No project folder until you pick one) before the first message: recent folders come from earlier panel conversations, recent Claude Code and Codex sessions and sites bound in Inspect & Ask; Choose a folder… opens the system folder picker (on Linux without one, type the path). The folder must exist and cannot be your home folder itself, a folder that holds others (such as /Users or /opt), or a system folder or anything inside one (such as /etc, /usr, /System or /opt/homebrew; on Windows, Windows and Program Files). It is fixed when the conversation is created and shown in its header. Pi conversations always use their private folder.

Only the agent works in the project. The panel's own files (its MCP configuration, attached images, and screenshots, downloads and recordings captured by browser actions) stay in the conversation's private folder and are never written into the project.

A repository page on a code host starts in your local checkout of it. On any page under github.com/<owner>/<repo> (a pull request, an issue, a file), and the same pages on GitLab (subgroups included), Gitee, Bitbucket, Codeberg and self-hosted Gitea, Gogs or GitLab servers, the panel looks through your recent folders for one whose git remote (origin, upstream or any other; https, ssh and git@host:owner/repo forms) is that repository. New conversations start in it, and the panel shows Matched by repository <repo> below the folder, with nothing to click. Self-hosted servers are recognized by your checkouts' remotes, with no setup. Remotes are read from .git/config (git worktrees included); git itself is not run. An ssh remote that uses a host alias from ~/.ssh/config (such as git@github-work:owner/repo) is matched by that alias's HostName (read only; Include is not followed).

Other sites that hold many projects go by project too, not by domain: App Store Connect apps (/apps/<appId>, TestFlight pages included), Vercel projects (/<team>/<project>), Netlify sites, Cloudflare Workers and Pages, Firebase and Google Cloud projects (the project id in the path or in ?project=) and Sentry projects. When a file in the root of a recent folder names that project, the folder is preselected, shown as Matched by project …: an App Store Connect app id in fastlane/Fastfile, Appfile or Deliverfile, the project name in .vercel/project.json or vercel.json, the name in wrangler.toml/wrangler.json, a project id in .firebaserc, or the project in .sentryclirc or sentry.properties. Netlify has no such file and relies on Remember. The rules are a table in the CLI; another site is one more row.

A site can also remember its project, in the same bindings Inspect & Ask uses. After you start a conversation in a different folder, the panel asks Use this project for <scope> from now on?, and the scope can be changed right in the question: on a repository page it is This repository (such as github.com/leeguooooo/image-use), on App Store Connect This app, on Vercel and the like This project, named by the page title ("Hellow (appstoreconnect.apple.com/apps/1234567890)"), with This path and Whole site as alternatives; on other sites it is the whole site, or just the current path. A site of many projects is never bound whole by default. A project named only in ?project= (Google Cloud, some Sentry pages) cannot be remembered per project, only by path, though it is still matched automatically. Remember saves it, and nothing is overwritten if that scope's binding changed meanwhile.

Which project wins: the binding with the longest scope covering the page; a project bound to the repository (or a path inside it) wins over the matched checkout; a broader binding (the whole site, an owner) does not override the repository's checkout and only picks among several checkouts. Existing site-wide bindings are kept. Review, change the scope of, or remove these bindings on the extension's options page under Sites bound to projects. A /resume conversation uses the resumed session's folder as its project.

Privacy and configuration

Recognized sensitive fields and secret-shaped strings are masked before new tool and approval records reach the local journal. Previously known secrets are scrubbed from later display text. This does not cover your own message text, image pixels, file bytes or the CLI’s own session files; secrets discovered later cannot erase older text. Result files remain with the conversation until it is deleted. See Privacy.

History stays on this computer. The chosen CLI may send tasks and page content to its configured model service, whose privacy policy applies. The panel adds no webpage content script; opening it can change the viewport. Agents & models discovers programs and previews configuration changes, with verified writes and restore that preserves later edits.

Skills and images

Type / to choose a skill, at the start of a message or after a space anywhere in it; ↑↓ move through the list, Tab or Enter picks, Esc closes. The list includes the skills of Claude Code plugins you enabled. Agents & models β†’ Skill library edits or enables library entries and imports local folders, ZIP/.skill archives or HTTPS Git repositories. Existing user/CLI skill folders are read-only here. Skills and assets are snapshotted before sending, so queue edits and explicit retries retain their revision. Panel-bound agents can list/read skills and save library instructions with chrome_use_skills.

Paste, drop or choose up to five PNG/JPEG/WebP/HEIC files with Attach images. The local CLI converts HEIC to JPEG with the system codec: sips on macOS, the HEIF Image Extensions and HEVC Video Extensions from the Microsoft Store on Windows, heif-convert or ImageMagick with HEIF support on Linux. The converted photo is turned upright and its EXIF metadata, including GPS, is removed. Images are reduced to a 2000-pixel long edge and at most 512 KiB each. They remain with the conversation, including queued edits and retries, and are removed with it. ACP agents must advertise image input. Attachments are sent to the configured model service.

Conversation options and recurring work

Conversation options overrides allowed tools and Codex/Claude reasoning effort; Codex also supports service tier. Save while idle; the next response uses the override without editing CLI defaults. Codex cannot remove its file-reading shell. OpenCode ACP cannot enforce built-in tool restrictions. The approval switch controls automatic answers within the chosen tool scope; Pi has no per-call approvals.

Agents & models β†’ Scheduled prompts supports intervals, daily/weekday local times and five-field cron, with agent, model, options, skill and explicit tool scope. Runs use independent background browser sessions, appear in history and notify on completion. Run now does not move the next scheduled time. Closing the panel does not stop the local service; stopping that service stops scheduling. Missed periods coalesce into one run after restart or wake. An active or unconfirmed run is never overlapped or replayed; inspect its history before acknowledging an unknown outcome.

When a panel task finishes, fails, stops, needs recovery or waits for your approval, Chrome shows a notification and the toolbar icon counts the unseen ones; clicking it opens the side panel on that conversation. The conversation the panel already shows in the focused window does not notify. Settings β†’ Notify when a task finishes turns it off.

Chats cannot schedule more tools or automatic approval than they have. Scheduled conversations can list schedules but cannot change or trigger them, or rewrite library skills for future runs. Agent completion does not establish that the requested page effect succeeded.

If Chrome rejects a panel turn notification, a later check in the same extension-worker session retries it. Disabled notifications and turns already being viewed remain suppressed. Viewing a turn while its notification is being created also consumes that turn if creation later fails, including across worker restarts.

A turn for each request

Requires CLI 1.5.215 and extension 0.5.58. Older extensions retain the previous view. Chrome Web Store submission, review and public availability are separate from the CLI release.

Each message has a card containing its attached page, browser steps, response, result files and turn outcome. Running steps expand; finished groups can collapse, with failed steps retained in the summary. Missing evidence in older history is labelled not recorded, never guessed.

Attached page and acting page

Sent with is the page attached when you sent the message. Acted on lists pages named by browser-action replies. An agent can open another tab, so these may differ. A cached target without an action reply is labelled Browser tools pointed at. A tab you merely activate is not recorded as a page the agent acted on.

What a passed check means

LabelMeaning
Agent's browser checks passedThe browser checks the agent chose for this request passed. Expand them to see what was checked; this does not guarantee the checks fully express or satisfy your request.
Some checks passedSome checks passed; others were not checked, failed or lack usable evidence.
Not checked / BlockedNo supporting check, or a blocker such as login, human verification or missing input.

Read-back, page assertions, landing URLs and download types establish only their specific checks. An existing file does not prove correct content, and a screenshot does not automatically prove completion. A finished turn, passed browser checks and fulfillment of your request are separate facts.

Screenshots, downloads and recordings

Result cards show file type, size and availability. Reads verify content; missing or changed files become unavailable. Existence, type matching and fulfillment of the request are shown separately. Delivery to the agent's requested destination is also reported separately from the conversation's retained copy.

Check Record this task before sending to permit recording for that message. Recordings move through capture, processing and availability states. A finished answer does not mean the recording is ready; missing encoders and processing errors are reported. Recording never starts automatically for progress display.

When an outcome is unknown

Stop requested, waiting for confirmation means interruption has been requested, not confirmed. This conversation's agent has stopped refers only to this conversation, not to other agents.

  1. Choose Review what may have happened to inspect possibly executed actions, last pages and result files. Do not immediately repeat a send, payment or submission.
  2. End the old agent process is offered only when a live process can be proven to belong to this conversation.
  3. After its exit is confirmed, Continue in a new conversation carries a reference to the old history. Creating the conversation does not resend the previous request.

No continue action is offered while process state remains unknown. After a service restart, some platforms cannot re-establish the old process tree's state and retain the unknown result.

Long tasks and continuous history

Repeated results, repeated failures or a long period without activity raise an advisory. It does not stop or retry the task. You can stop, wait or inspect the steps. Context compaction leaves a marker while history remains available; it never turns an unchecked action into a successful one.

Click Pick element at the start of the composer toolbar (or press Alt+Shift+C on the page) and click elements on the page: they wait as chips in this tab's side-panel composer and go to the agent with your next message. A code-change request can also go to an existing local terminal session. See Inspect & Ask.