Grok Build isn't the coding agent most people run. Plenty of agents have more users. It still got PonyMux support next, for one reason: it can search X.
PonyMux 0.11.3 supports the Grok Build CLI, the grok command, the same way it supports Claude Code and Codex. Each Terminal running Grok shows its live status. Approvals mark it Waiting for you. Its conversations show up in search, and you can resume, fork, or hand them off to another agent. The Inspector shows its model and context, and the usage card shows where you are in your billing period.
Why Grok, and why now?
A lot of what I want to know about my tools never makes it into the docs. A library ships a breaking change and the first reports are on X. An API starts returning odd errors and someone has already posted the workaround. A new model comes out and the useful takes are in replies, not in blog posts.
Grok Build has an X search tool built in. It can look that up in the middle of a task, next to reading files and running commands. Claude Code and Codex can search the web, but neither ships a way to search X.
That made Grok the agent I actually wanted in my Terminal list, not just the next one on a list sorted by users.
Not the biggest agent. The one that can read X.
What works with Grok in PonyMux?
Everything that already worked for Claude Code and Codex:
- Status on the Terminal row: Running, Waiting for you, Unread, Standing by, Will come back, Error.
- Approvals that mark the Terminal as Waiting for you and clear as soon as you answer.
-
Search across your Grok conversations with
⌘K, including the feedback you type when you reject a tool call. - Resume and fork from the Inspector or the Sessions panel.
- Handoff to Claude Code or Codex, and from them to Grok.
- The Inspector, with model, reasoning effort, and context.
- Usage on the Grok card in Agent Integrations.
Subagents count too. When a Grok subagent asks for approval, the main session's Terminal shows it.
How does PonyMux know Grok is waiting for you?
Grok sends a hook event when it asks for approval, but nothing when you answer. Without that second event, the Terminal would stay on Waiting for you after you'd already said yes.
So PonyMux reads the key you press in Grok's approval menu. Pick an option with a number key or Return, and the status changes right away.
What does the Inspector show for a Grok session?
Grok writes more about the session than the other two agents do, so the Inspector shows more. Next to the model and reasoning effort, the context line reads as used out of total, like 23k / 256k in context. For Claude Code it's only the tokens in use, and Codex sessions show no count.
The Session section has the same actions as the other agents: Resume, Fork Session, and a Handoff to… menu that lists the other two agents.
Can I hand a session from Claude Code to Grok?
Yes. Handoff works between all three agents, in any direction. Pick Handoff to… in the Inspector and choose the agent. PonyMux writes a handoff file with the session's summary and last five exchanges, then starts the other agent in the same folder with a prompt to read it. The original session stays as it was, so you can still resume it.
Where does the Grok usage number come from?
The Grok card in Agent Integrations shows one bar for your current billing period, labeled 7d or 30d. PonyMux reads the numbers Grok last wrote to its own log, so they're as fresh as your last Grok turn. It doesn't read your Grok credentials.
How do I set it up?
Install Grok Build from x.ai/cli, then select the link button at the bottom of the PonyMux sidebar and click Connect on the Grok card. PonyMux asks before it writes anything, and for Grok it adds a single file, ~/.grok/hooks/ponymux.json, without touching your other hooks. Start a new grok session in a PonyMux Terminal and its status shows up on the row.
For the details on each provider, see Agent integrations.
Top comments (0)