Solutions
Answer a blocked agent from your phone
Two things fix it. The app has to show which agent is waiting, and answering has to work from a browser or phone, not just the machine running the agent. Forkbench's Thread flags whichever pane is blocked, and Talk lets you answer from anywhere, for Claude Code, Codex, Gemini CLI or Aider alike. If you only ever run Claude Code, Anthropic's own Remote Control, included on paid plans, does the same for that one agent.
An agent working unattended does not fail loudly. It stops, asks a question, and waits. If you walked away twenty minutes ago it has been idle for twenty minutes, and if you are running four of them you have to click through every tab to find out which one it was. The cost is not the question, it is the dead time between the agent asking and you noticing, and it lands hardest on exactly the long unattended runs an agent is most useful for.
Compare what leaves the machine before you pick, because every option here sends conversation text to a server to render it. Anthropic documents that while Remote Control is connected, the session transcript including your messages, Claude's responses and tool activity is stored on their servers. Talk sends the readable text of the live conversation and its title, never raw tool output or terminal scrollback, and drops what it held a few minutes after the session stops being live unless it is on a shared board. Neither is end-to-end encrypted, and any page telling you otherwise is worth distrusting. The design constraint worth knowing is that Talk is attach-only. It joins a conversation that is already on screen; it cannot start one, and there is no way to spawn an agent remotely from a phone. That is deliberate, because a remote surface that can start work is a remote surface that can start work you did not ask for. What you get is the ability to unblock, redirect and queue an instruction into something that is already running. The machine has to be awake for that, which is what Sleep Control is for: it keeps the machine running with the lid shut so a long job finishes and stays reachable.
How it works
- 1Leave the agent running in its Thread. When it stops for an answer, the Thread shows which pane is waiting rather than making you check each one.
- 2Open Talk in any browser, on a phone or another computer, and pick the conversation you want.
- 3Read what it asked and reply. The answer routes back to that terminal on your machine and the agent carries on.
- 4Queue the next instruction while you are there, if you already know what you want it to do next.
- 5Turn on Sleep Control before you leave, so the machine stays awake with the lid closed and the session is still there when you look.
Straight about the guarantee: Be clear about what crosses the network here, because it is different from the rest of the product. To render a conversation in a browser and route your reply to the right terminal, the readable text of that live conversation and its title pass through us. Raw tool output and terminal scrollback are never sent at all, and a conversation that is not on a shared board is deleted a few minutes after it stops being live. Your notes and your Vault secrets are sealed under keys we do not hold and never work this way. Talk is free.