Angelia · Updated 27 October 2026

Angelia for Grok Build: setup, approvals and limits

Angelia connects Grok Build, xAI's coding agent CLI (grok), on your Mac to WhatsApp and Telegram. Each chat gets its own Grok Build session, in its own folder. Grok Build reads the profile's CLAUDE.md and Claude settings, so the instructions and rules you wrote for Claude Code work here too. Angelia runs no model of its own: Grok Build does the work, under your own grok login.

What Angelia does for Grok Build

• One session per chat. Each chat is routed to a profile: a folder with its own instructions and memory. Angelia keeps one grok agent stdio process per chat, and the session comes back after a restart. /new in one chat never touches another.
• Your instructions and rules. angelia compile writes the profile's deny rules into its .claude/settings.json, which Grok Build reads. Grok Build reads a folder's CLAUDE.md and settings only once you trust the folder. The wizard, onboarding and /backend grok do that for you.
• Grok Build's own permission modes. The profile's permission_mode goes to Grok Build as its own mode. Its questions come to the chat.
• It keeps running. As a LaunchAgent, Angelia starts at login and comes back after a crash. The chat answers whenever the Mac is on and awake.

What you need

• A Mac with Node 22.23.4+, 24.22.1+ or 26.12.1+ and git.
• Grok Build, signed in with grok login. Measured with 1.0.13 to 1.0.40; check with grok --version.
• A chat app on your phone. For Telegram, a bot token from @BotFather (more). For WhatsApp, a second, established number, not your personal one (why).
• Optional: brew install openai-whisper ffmpeg for voice notes.

Set it up

curl -fsSL https://useangelia.com/install | sh
angelia service install

The first line checks the release's signature, installs Angelia and starts a setup wizard (read the script). Pick Quick or Advanced, then Telegram, WhatsApp or both. Pick Grok Build as the CLI. The wizard asks to mark the profile's folder trusted in grok: say yes. Then send the bot (or the second number) a message from your phone to pair the chat. The second line keeps Angelia running.

Already running Angelia with another CLI? An owner sends /backend grok in one of the profile's chats. Angelia edits the routing table, ~/.angelia/workspace/routing.yaml, with a backup, compiles the profile for Grok Build, trusts its folder in grok and starts a fresh session.

If you set the profile up by hand, run this once in its folder, or Grok Build will not read its CLAUDE.md. It trusts the folder, then prints Project trusted: yes:

grok --trust inspect

What works from the chat

Approvals

Grok Build decides by itself which commands to ask about. Measured in September 2026 (grok 1.0.13 and 1.0.30): touch ran by itself; rm, git init and curl asked. The question comes to the chat as it does for Claude Code. bypassPermissions approves everything without asking.

Answering a Grok Build question from the chat works from Angelia 0.3.23. In 0.3.22, a bug keeps yes <id> from reaching Grok Build's questions. The fix ships in 0.3.23. Run angelia update once 0.3.23 is out.

No sandbox

Angelia has no sandbox for Grok Build: sandbox: true is for Claude Code, Codex and pi, and compile says it is ignored here. The deny rules stop Grok Build's file tools, not a program its shell runs. So keep the shell asking in any profile other people can talk to.

• Grok Build reads a path rule only as a full path. A rule written with ~ is ignored, and compile warns about it.
• MCP servers set up for your whole user reach every folder, so compile denies each one the profile was not given. Compile does not write the servers a profile was given yet: add them with grok mcp add --scope project in the profile's folder.

Voice notes, photos and files

A photo or document you send lands in the profile's .inbox/ folder, up to 20 MB, and Grok Build gets a [file: path] line. A voice note arrives as [voice note: path], and the agent runs angelia transcribe <path>. Grok Build sends files back with angelia send-media. Every path is checked first: never a credential location, never another profile's folder.

Chat commands

/new, /stop, /status, /resume, /model, /effort, /backend, /restart and the others /help lists are answered by Angelia, without spending a token. /model alone shows the models grok models lists. Any other slash command is passed to Grok Build as text. Long answers are split as for every CLI.

Tested

On 2026-10-27, with Angelia 0.3.22 and Grok Build 1.0.40 on macOS, in a WhatsApp group routed to one profile in permission_mode: default:

• An answer came in about 11 seconds. A long answer arrived in 2 messages, in order.
• /status, /new and /stop work.
• Grok Build's questions could not be answered from the chat (ids starting call-). Fixed for 0.3.23.

Grok Build on Telegram is not tested yet. Voice notes are tested with Claude Code only.

Limits

• macOS only for now. Linux is next. The Mac must be on and awake: while it sleeps, nothing answers.
• Approvals from the chat need Angelia 0.3.23 or newer.
• No sandbox, as above.
• angelia ask (one profile asking another's agent) does not reach a Grok Build profile yet. Give it a task with angelia turn instead.
• API key variables (XAI_API_KEY, GROK_CODE_XAI_API_KEY) are kept out of the agent's environment.
• Prompt injection is not solved. A web page or a forwarded message can still talk the agent into what its mode allows. Read the security model before you give the bot to anyone but yourself.
• It is a young project (v0.3), maintained by one person.

curl -fsSL https://useangelia.com/install | sh