Claude Cloud Added in 0.9.7

A Claude Cloud chat runs Claude Code in a sandbox at Anthropic instead of on your relay host. The sandbox works on its own clone of a GitHub repository. It sees what is pushed to GitHub and nothing on your machine, and its results come back as branches the agent pushes. The chat is billed to the Claude account the relay host's Claude CLI is logged in to.

In OAR it is a conversation like the others: the reply streams in, questions arrive as cards, and it works from your phone.

What it needs #

• The Claude CLI on the relay host, logged in with a claude.ai account (Settings → Providers → Claude → Relogin, or run claude on the host). The same login serves Claude chats.
• GitHub connected to that Claude account: claude.ai/connect-github.
• The Claude GitHub app allowed on every repository a cloud chat should work on (GitHub → Settings → Applications → Claude → Configure).
• An environment on the Claude account. An account without one gets its default one when you open claude.ai/code once.

Switching it on #

Open Settings → Providers → Claude Cloud.

The tab also shows the account the chats are billed to, and where the relay reads the login from and until when it is valid.

Switching the provider on is your consent to the use of the Claude login described above:

• The relay and its cloud workers only read the access token. It is sent to https://api.anthropic.com and to no other host.
• It is held in memory only. It is not written to the database or to a log, not sent to the browser and not put into an error message.
• OAR never refreshes it and never uses the refresh token. The CLI keeps its own login fresh. When the token has run out, the relay reads the file again and, at most once in five minutes, has the CLI check its own login. A login that is still expired ends the turn with a note to log in again.
• While the switch is off, the relay asks Anthropic for nothing with this login: it lists no environments and reads no account usage.

Switching Claude Cloud off takes it out of New Chat. Cloud conversations that already exist keep working.

Starting a cloud chat #

Choose + New Chat → Provider → Claude Cloud. The Working directory stays, but here it only fills in two fields, which you can also pick from a list or type yourself. There is no reasoning effort to choose. The cloud session itself is created by your first message. The dialog is described in Starting a conversation.

Repository #

Read from the folder's origin remote, or the remote of its upstream branch. With the field focused, OAR lists the repositories the Claude GitHub app can reach for the account. The list is read from Anthropic and kept for a minute; ↻ next to the label reads it again. If it cannot be read, the reason is shown under the repositories used before. Repositories used in earlier cloud chats on this relay come first and are marked "used here"; other marks are "private", "public", "archived" and "no access". Typing filters the list by owner/name; the arrow keys and Enter pick, Escape closes it.

You can also type or paste owner/repo or a GitHub URL, https or ssh. Only GitHub repositories are accepted.

Branch #

The folder's current branch; a repository picked from the list brings its default branch. With the field focused, OAR offers the repository's branches as the relay host's git sees them (git ls-remote, with the host's own GitHub login). When the host cannot read them, the field is a plain text field. Leave it empty for the repository's default branch.

Warnings #

Under the fields OAR warns about what the cloud clone will not have:

The first two warnings only appear while the fields still name the folder's own repository and branch.

If Start chat is refused, an error line under the fields says why: for example Enter a GitHub repository., Not a GitHub repository. Use owner/repo or a github.com URL., Not a valid branch name., Claude Cloud is not enabled on this relay, or No cloud environment is set. with the advice to pick one in the settings.

Working in a cloud chat #

The reply streams in as the agent writes it, with thoughts, tool activity and subagent bubbles. While the sandbox starts, the tool activity shows its steps as lines starting with Cloud:, then Claude Code running in the cloud with the model.

• The cloud line. Above the composer: the repository, the branch, a claude.ai ↗ link to the session, one ⇡ branch link per branch the agent pushed (the newest 20), and what the session has cost so far as Anthropic reports it. A pushed branch opens GitHub's comparison with the branch the chat started from (the branch itself, when the agent pushed to it).
• Open on claude.ai. ☁️ Open on claude.ai in the chat's ⋯ menu opens the session on claude.ai, once the session exists.
• Questions. The agent's questions (AskUserQuestion) arrive as question cards, multi-select included. A tool that asks for permission gets an Allow / Deny card. A card you answer on claude.ai closes in OAR.
• Stop. Stop interrupts the cloud turn. If the cloud does not confirm within 30 seconds, the turn ends in OAR with a note that the agent may still be working there, with the link to the session.
• Images. JPEG, PNG, GIF or WebP, up to 5 MB each, sent inline. Other attachments are refused: the sandbox cannot read the relay host's files.
• Follow-ups. A message goes into the same cloud session, also after it has been idle for a long time. A message sent during a turn waits for the turn to end.
• Changing the model. The model can be changed between turns in the composer, as in a local Claude chat. The sandbox switches before the next message.
• Commit attribution. Commits and pull requests the agent makes in the sandbox end with the lines the Claude tab's attribution setting says; see Commit attribution. The commit's author is still the sandbox's git identity (Claude).
• Background work. A turn stays open while the agent's background work (a subagent, a long command) runs, so the closing reply arrives in the same turn.
• Turns the session starts by itself. When the agent's background work outlives the turn, or when you send a message on claude.ai, the answer appears in the chat as a continuation reply, with a line saying where it came from.
• Archiving and deleting. Archiving or deleting the conversation archives the cloud session at Anthropic; unarchiving brings it back. Nothing is deleted there.

When the browser or the relay is away #

The agent keeps working in the cloud. A restart of the relay or of the conversation's worker in the middle of a turn does not send your message twice: the new worker finds the turn in the session's event log and follows it, and a turn that finished in the meantime is answered from the log. A dropped connection to the session is reopened where it broke off. See Restarts and resume for the same in general.

At the usage limit #

A turn the cloud refuses at the Claude usage limit is paused like a local Claude turn: a note with the reset time, the banner, and the work carries on by itself after the reset. See When a usage limit is reached. The warning before the limit is a line in the tool activity, Claude usage limit warning (...), not a banner.

What a cloud chat does not have #

Nothing from the relay host reaches the sandbox.

• Relay tools. No preview, no relay_tool, and no media embedded by host path. See By provider.
• Folders. No working directory to change (Change CWD is missing from the ⋯ menu) and no Session folder to browse.
• Relay modes. No relay modes, no reasoning effort and no context size: the three selectors are hidden.
• Steering and tasks. No mid-turn steering and no background task panel.
• Attachments. Only images.

Usage #

With Claude Cloud on, Check Usage has a Claude Cloud card:

• a meter per dollar credit the account holds, with what is used, what is left and when it expires;
• Cloud spend (OAR sessions): the cost Anthropic reports for the cloud sessions this relay still has, with this conversation's own cost when you open the modal from a cloud chat;
• the latest cloud session report: its cost, and cost and tokens by model.

These figures are Anthropic's reported cost, not a billing statement, and sessions started elsewhere are not counted. A credit that is offered but not yet claimed is mentioned with a link to claude.ai. Cloud turns spend the cloud credit first; without a credit they count against the Claude limits.

The Claude card in Check Usage also reads its limits live from the account while Claude Cloud is on. The 🧠 button shows how much of the session's context window is used, without categories. No usage line is attached to the replies.

Starting a cloud session from an agent #

An agent can start a cloud chat with create_session, using provider: "claude-cloud", a repo and optionally a branch; see create_session. It works for an agent on a paired relay (under the rules for starting any session from there) and for an agent on this relay once agent sessions are switched on. The same checks apply as in New Chat: the provider is on, an environment is set, the repository is on GitHub. A chat without a requested model starts with the tab's default model. An agent can also read, prompt, wait for and stop a cloud chat you created.

Environment variables #

If it does not work #

If Claude Cloud is missing from New Chat, enable it in Settings → Providers → Claude Cloud.

When a failure has a known cause, the reply is a system note that says what happened and what to do. Then send the message again.

A connection problem that may pass does not fail the turn: the message goes back into the queue and is delivered again.