Agent Auto-Continue

How it works

1. 01

   Pick a thread

   Conversations from T3 Code, Claude Code, Codex, OpenCode and the Claude and ChatGPT desktop apps, in one list, most recent first.

2. 02

   Write the nudge

   Usually just "Continue". Anything you would type into the thread yourself works.

3. 03

   Choose when

   Quick picks for +5 min, +30 min, +1 hour or tomorrow morning, an exact date and time, or as soon as the usage limit resets.

Built to never double-send

Sending a message to an agent twice is worse than not sending it at all, so the app is careful.

• Re-checks the thread before sending, and cancels if it is gone, archived, or you already replied.
• Never retries an interrupted or ambiguous send. It marks it unconfirmed and only reads the thread to check.
• Keeps its queue through quitting, restarts and sleep, and catches up on anything it missed.
• Reaches each agent through its own local interface. No remote service of its own, no simulated keystrokes, no changes to the agent apps.

Also built in

• Keeps going on its own: up to a turn limit, until you stop it, or until the agent says a stop phrase such as "TASK COMPLETE".
• Skips a send while the agent is still at its usage limit, and can start the moment the limit resets.
• Optional remote control from a phone or another agent, through a token-protected API and MCP server, on this Mac or over Tailscale.
• Can keep the Mac awake while queued work waits or runs.

Guides

Short pages on the things people actually search for, plus the docs.