Skip to content
Moontors

Moontors 0.7.6

Released October 24, 2026.

Moontors, the agentic development environment (ADE) for Claude Code, Codex and Gemini CLI, now runs Claude Code 2.1.309, Codex 0.164.0 and Gemini CLI 0.65.0, with what each of them added. What you send while the agent works waits in a queue and goes as soon as it can, the header counts the sessions that need you, and open files follow the agent’s edits as it writes them.

What’s new:

  • Send whatever you like while the agent works. A message it can take goes into the turn; everything else waits in the session’s queue, in the order you sent it, and goes as soon as the turn before it ends: a message it couldn’t take, every command, the agent’s own and Moontors’, and /rewind, /branch, /cd, /add-dir, /subtask, /recap and /setup-default-sandbox, whose pick waits with them. Once anything waits, Send says Send when the turn ends, and the / list marks what Runs when the turn ends. Each waiting row has Edit, which takes it back into the message box, and Discard. Stop ends the turn and gives nothing back: what you’d sent goes first, then the queue. A message sent into a turn shows last, “Waits for the agent to take it in”, until the agent takes it, then where it did. On the computer and the phone. See Sending while the agent works.
  • Where Claude Code’s own terminal would stop to ask what to do next, a card asks you: when a model refuses a request another model can take, Retry on that model or Edit the prompt, which puts it back in the message box, the reply it had begun folded away; when a model goes on only on your usage credits, Use usage credits or Switch to the default model, with what you have left; and auto mode’s notice that its checks are billed to you, Continue in auto mode or Stop. On the computer and the phone. See Choices.
  • A call Claude Code’s Auto mode turned down that one of its rules can name exactly, a command, or a file already there read or written by its full path, offers Allow for this session beside Keep denied; a folder, or a file not written yet, gets none, as Claude Code’s rule for it would cover more. It adds that rule to the session’s rules, where you can remove it, and the agent runs the call again: in the same turn while it runs, or as your next message while it’s idle or something waits. On the computer and the phone. See Allow for this session.
  • The header counts the sessions in each state that needs you or that you’re watching, “2 running”, “1 waiting”, “1 action denied”, “1 error”, across every project, each session once, as its card shows it: running while a turn runs or its background work goes on, and waiting whatever else it runs. Click a tag to go to the next session in that state, round to the first. On the phone the same tags head the board, in place of the project filter, and a tap shows the next session’s card on the board, without opening it. See The header.
  • The phone’s board shows its sections as tabs over pages: tap a tab or swipe between them. Drag a card by its handle within its page, or rest it at the screen’s edge to carry it to the section beside; drag a tab to move its section. See The board.
  • With Claude Code, the Status tab shows What fills the context, part by part, each MCP server, instruction file, skill and subagent with its tokens and share, and /context opens it there. What uses your limits gives Claude Code’s scan of your sessions over the last 7 days or 24 hours, its advice with each part’s share, and Fix, which applies its fix to the session: the effort, and the rest as one message, or Fix when the turn ends while the agent works. A session’s usage also names the part of its tokens read from the cache and the part spent thinking, where the agent counts them. See Context, usage and limits.
  • Codex sessions have Fast mode, beside the model picker, and /fast, as Claude Code’s do, on a model that offers it. /btw asked while a Codex session is idle is answered on the conversation itself, so the answer knows all of it. See The model picker.
  • /feedback, with Codex, opens Report on this session: say what it’s about, add your words and choose whether Codex’s logs and the conversation go with it, then Save… it as a ZIP where you choose, or Send to OpenAI…, which lists everything that goes, and from whose account, before you press Send. See The built-in commands.
  • Codex and Claude Code sign in to Amazon Bedrock: choose Sign in with Amazon Bedrock, then an AWS profile or your environment’s credentials, or give a Bedrock API key or AWS access keys, and check the AWS Region. Your keys are never shown again, and Accounts shows the sign-in as Amazon Bedrock. See Ways to sign in.
  • An image Codex makes is an Image row, titled by the prompt it was made from, which opens to the picture, on the computer and the phone; Moontors keeps it with the session, as it keeps a file you send. Where Codex’s image limit stopped it, the row says when the limit resets. A hook of Codex’s that fails, blocks or stops the agent shows as Claude Code’s does, in Codex’s words, and a ChatGPT plan reads by the name Codex gives it, as Pro 200 or Academic Researchers. See Tools and subagents.
  • A folder added with /add-dir brings its own Claude Code commands and skills into the / list and the skills list. On Linux, Claude Code’s sandbox has a fourth switch, Allow nested user namespaces, off by default. Where Claude Code reads pages through a subagent of its own, as it does for some accounts, the fetch shows inside that subagent’s row, linked to its page. See /add-dir.
  • Gemini CLI’s edits and new files are written through Moontors, which refuses a write that would lose what isn’t text and says why in the transcript. Gemini’s settings list hides a secret’s value, as a token in its telemetry headers. See Tools and subagents.
  • Where your organisation’s managed Codex settings give new projects a folder, Add project’s picker starts there. See Projects.
  • A / after a space anywhere in a message opens the commands list too, with your own commands and skills the agent still runs when named mid-sentence. See The commands list.
  • Files open in Moontors follow the agent’s edits as it writes them, not only at its turn’s end: a file with no unsaved changes of yours updates, one with yours says it changed on disk, and a file the agent makes shows in the tree. See Editing beside the agent.
  • A code block in a reply has Copy code, shown as you hover it, and sits at the message’s own padding; on the phone, a copy icon at its top right. See What the transcript shows.
  • A question card’s Decline sits on its first row, after its tabs, so the question starts right under it. The Git view’s file lists show their scrollbar whenever they can scroll. The terminal’s Clear keeps the whole line you’re typing on where it wraps. See Git.
  • On the phone, Android’s Back goes back one step, from a session to the board and through pairing’s steps, never out of the app but from the board. Pairing by QR code says the phone will ask for the camera, then Continue brings up the phone’s own question; with the camera off, Open Settings, or enter the code instead. Pairing by code says it can’t reach a computer that doesn’t answer within 15 seconds, and Try again asks for the code again, and a paired computer that moves to another network is found again. See Pairing.
  • Fixed: a compaction is one line in the transcript, however often the agent says it’s still compacting.
  • Fixed: the page a device-code sign-in names opens when you click it.
  • Fixed: a command Claude Code runs in its own context shows its steps as they happen in a fresh install’s first turn too.
  • Fixed: a terminal tab whose terminal has already gone closes without an error.
  • Fixed: a mode or model picked while a session was being created is no longer lost; the dialog holds its choices until the session’s made.
  • Fixed: a model switch Claude Code confirms slowly no longer reads as no answer, as Moontors now waits up to 20 seconds for it.
  • Fixed: Gemini CLI’s tokens out include its thinking, as Claude Code’s and Codex’s do.
  • Fixed: the first pairing on an iPhone asks for the local network first, once you choose Scan the QR code or Enter the code, and waits for your answer instead of failing; refused, Scan the QR code still pairs through a tailnet. See Pairing.

After this update, Codex may ask you to sign in again to an MCP server that signs in with OAuth: the server shows Sign in, which opens its sign-in page in your browser. See /mcp.

Claude Code runs on Amazon Bedrock in Moontors through its sign-in alone: a CLAUDE_CODE_USE_BEDROCK your shell sets no longer reaches it. If you used Bedrock that way, choose Sign in with Amazon Bedrock in Claude Code’s sign-in. See Ways to sign in.

Moontors for iPhone is offered in every country and region but China mainland.

The Windows and Linux builds still aren’t signed, so Windows asks once before Moontors opens. The README’s “First launch” and “Updates” sections say what to expect on each platform.

Found a problem, or have an idea? Open an issue: https://github.com/moontors/moontors/issues/new/choose

Downloads

See First launch and Updates.

© 2026 Moontors. All rights reserved.

Moontors is an independent agentic development environment, not affiliated with or endorsed by Anthropic, OpenAI or Google, or the makers of Conductor, Emdash, Orca, Paradev (Superset), Vibe Kanban, opcode (Claudia), Cursor, VS Code, OpenCode or any other tool this site names.

Google Play and the Google Play logo are trademarks of Google LLC. Apple and the Apple logo are trademarks of Apple Inc., registered in the U.S. and other countries. App Store is a service mark of Apple Inc., registered in the U.S. and other countries.