Plan mode, a live task list and email sign-in

Multilo 2.0.145 is out for Windows and macOS. It changes how you start a big task, how you follow it while it runs, and how you sign in.

Plan mode

Ask for a plan before anything changes. In plan mode the assistant reads your files and looks around, then proposes a plan. Nothing is edited until you approve it.

• Turn it on with Shift+Tab in the chat box, by choosing Mode → Plan in the chat box's settings menu (the sliders icon), or by typing /plan. A Plan mode chip in the chat box shows it's on. Click the chip, or press Shift+Tab again, to leave.
• It stays on until you approve. Reply to the plan with "drop step 3, keep the Methods" and the assistant revises the plan instead of starting work.
• Edit the plan yourself. Click Edit on the plan card, change what you like, then click Approve & run. Where your version differs from the assistant's, yours wins.
• It can look before it plans. Read-only commands such as ls, grep or git log run as usual. Any other command asks first, in every mode, Bypass permissions included.

A live task list above the chat box

On longer jobs, the assistant's to-do list now appears above the chat box as Tasks, with a count such as 1/3 done. Finished items are ticked and struck through, and the one in progress is shown in bold. Click the header to fold the list to one line that names the current task. Once every item is done and the run has finished, the list clears itself.

Sign in with Google or email

The sign-in screen now offers two quick ways in:

• Continue with Google goes straight to Google's account picker in your browser.
• Continue with email stays in the app. Enter your email and Multilo sends you a sign-in link. Click it on this computer and you're in. Reading your email on another device? Enter the 6-digit code from the same email instead.

If you're creating an account, click the link in the email the first time. Other ways to sign in still opens the website's sign-in page.

Long chats that keep going

When a conversation gets close to the model's context limit, Multilo summarises the earlier part so the chat can continue. You can now see when that happens, and control it.

• A Compacted conversation line appears in the chat where it happened, with the size before and after. Click it to read the summary.
• Hover the context meter by the chat box to see where auto-compaction will happen. Once the context is 75% full, it offers Compact now.
• Type /compact to compact whenever you like, optionally followed by what to keep, such as /compact focus on the methods.
• Settings → AI Agent has an Auto-compact switch and an Auto-compact at setting. By default it compacts at 90% of the model's context window, and a number you set can only make it compact earlier.
• If a request turns out to be too long for the model, Multilo compacts the conversation and sends it again by itself, instead of failing.

Approvals that fit how you work

The approval card now has four choices, on keys 1 to 4:

• Allow runs it once.
• Allow "…" in this chat stops asking for the rest of this chat. It's forgotten when you close the app.
• Always allow "…" in your project stops asking in every chat in that project, and nowhere else.
• Decline.

Also:

• An approval or plan card now waits for you, however long you're away. Before, it declined itself after five minutes. Stop closes it.
• In Autopilot, Multilo now asks before a tool from a connected app (MCP server) writes to an outside service, such as sending an email or creating an issue. Tools that say they only read still run without asking.
• Autopilot also asks before reading files outside your open projects.

Find a chat by what was said in it

Search sessions (Ctrl+K, or ⌘K on a Mac) now matches what was said in your chats, not only their titles. Each result shows the line that matched, so you can tell at a glance which chat found that dataset.

Edit, retry and rewind keep your attachments

Edit and resend, Run this message again, Try again and Rewind → Conversation only now bring back the files, folders and images you attached, not only your words. Files are read again from disk, so you get their current version. An image pasted from the clipboard can't be brought back, and Multilo tells you which ones.

Esc to stop

Press Esc to stop the assistant mid-run. Work that's already done is kept, and the assistant is told you interrupted it, so it checks where things stand before carrying on. If an approval card is open, Esc declines that card instead. The Stop button's tooltip now shows the shortcut.

Clearer credit and plan messages

When your plan is what's stopping a reply, a notice above the chat box now says what happened and offers the button that fixes it:

• You're out of credits, with the date your credits reset.
• You've used your included credits, with Turn on extra usage.
• You've reached your monthly spending limit, with Raise limit.
• Your plan is paused, with Resume plan.
• You've reached today's limit, with the time it resets.
• This model isn't included in your plan, with the choice to upgrade or pick another model.

You also get a heads-up at 75% and 90% of your credits, with the reset date. If Auto is choosing lighter models to make your last credits go further, or a model you picked is no longer in your plan, the notice says so. And you won't see an Upgrade button if you're already on the top plan.

Smaller changes

• When a model declines a request because its safety system flagged it, Multilo now says so and tries another model once, instead of reporting an empty reply.
• Background commands, such as a long analysis or a build, no longer stop after 30 minutes. They run until they finish or you close Multilo. Type /tasks to see what's running.
• A chat you reopen, or restore from the archive, comes back with its whole conversation, so the assistant still has your first request. Archived agent sessions are deleted 30 days after you archive them, and the Archived view says so.
• Links in a document to a file on your computer now open that file, and links with a folder or drive letter in them survive a save.