Changelog
What's new in Moxie.
New features, improvements, and fixes in every release.
Claude Code
Codex
DeepSeekSwitched to Codexnow
Claude Code reached 97% of its limit, so new requests go to Codex.
Back on Claude Codenow
Claude Code has usage left again, so new requests go to it instead of Codex.
Introducing
Smart Router
Smart Router keeps Claude Code working when an account runs out. Line up your accounts in a router, and Moxie watches the one in use. When it reaches the switch threshold (97% by default), new requests move to the next account in the chain, before the old one hits 100%. When an earlier account's limit resets, Moxie moves back to it.
Each account in the chain can serve its own models, so a Claude account can hand over to Codex or Gemini. Open Smart Router in the sidebar to make your first router.
Read more- New
Watch it route. The Smart Router page shows your chain live: which account is serving, which one is next, and how close each is to its limits. Home shows when a router is choosing your account.
- New
Credits as a last resort. When every account in the chain is at its limit, Smart Router can move to one that still has Codex credits or Claude extra usage. Turn it off on the Smart Router page.
- New
Your pick stays put. Choose an account yourself on Home or in the menu bar, and Smart Router leaves it alone until it runs out.
- New
Know when it switches. A notification tells you each time Smart Router moves to another account, and when every account in the chain is at its limit.
- Improved
Presets is now Slots. The sidebar page for your model presets is now called Slots, so it isn't confused with Smart Router's routers.
- Improved
Limits are noticed sooner. Moxie now reads an account's usage as soon as a reply starts instead of when it ends, so limit alerts and the Dock gauge keep up during long replies.
- Improved
Rate limits on Codex and OpenAI accounts come through properly. Claude Code now gets them as rate-limit errors, and Moxie's usage limits pick them up.
- Fixed
A brief rate limit on an API key account no longer leaves it marked as rejecting requests for the rest of the session.
- Fixed
Choosing an account while a Foundation Model account was still starting no longer gets overridden once it finishes starting.
This update fixes the VS Code extension, which 2.3.0 could not install.
- Fixed
The VS Code extension installs again. Moxie could not download its VS Code extension, so Claude Code's model picker in VS Code, Cursor, and Windsurf did not follow your account and preset switches, and Update now in Settings showed "HTTP 404". Moxie now installs the extension on its own after the update. Reload any open editor window to start using it.
Introducing
Model slots
Claude Code's model picker has four slots: Fable, Opus, Sonnet, and Haiku. Moxie now fills each one with a different model from your account, so your strongest model and your fastest are one /model away, whatever the provider.
Pick a preset on Home to fill all four at once. Claude, OpenAI, Gemini, and DeepSeek accounts come with one ready to go. To change a single slot, choose Edit slots on Home, or make your own presets in Settings > Presets.
Read more- New
Presets for every provider. Make your own presets for any account, including OpenRouter, Ollama, and OpenAI-compatible ones, on the new Presets page in the sidebar, then pick them from Home.
- New
VS Code keeps up. Claude Code's model picker in VS Code, Cursor, and Windsurf now follows your account and preset switches, through a small extension Moxie installs for you. See the "Using with VSCode" guide.
- New
See which apps are busy. Click "Active now" on Home to see which apps, such as Terminal, VS Code, or Claude Desktop, are sending requests.
- New
A Dock bounce when limits change. While Moxie is in the background, its Dock icon bounces when your active account hits its limit or frees up again. Turn it off in Settings > Notifications.
- Improved
New OpenAI, Codex, and Gemini accounts start on a ready-made preset.
- Improved
The free trial starts without an email, and Moxie no longer needs Claude Code to get started.
- Improved
Home shows the Claude Code update banner only when the update brings new models.
- Fixed
Usage meters for Claude accounts no longer jump between two readings.
Claude Desktop can now chat through Moxie, and sessions with many MCP tools stay fast.
- New
Claude Desktop works with Moxie. Turn on "Configure Claude Desktop" in Settings > Server and the Code tab in Claude Desktop chats through your active Moxie account, with its models in the picker. Turn it off to go back to your previous setup. The website has a "Set up Claude Desktop" guide.
- New
Setup on first launch. After the trial starts, Moxie offers to set up Claude Code and Claude Desktop for you.
- Improved
MCP tools no longer weigh sessions down. Claude Code now loads a tool only when it needs it, with every account. Sessions with several MCP servers stay fast, use less of your plan, and stop summarizing over and over. Restart any open Claude Code session to pick this up.
- Improved
New models show up right away. Claude, Codex, and Gemini accounts now always see the newest models, even when the apps on your Mac are older or missing. If a brand-new Claude model needs a newer Claude Code, Home warns you to update.
- Improved
Deleting an account or preset, restoring a backup, and removing a license now confirm with a native Mac alert.
This update fixes a bug that could make Claude Code summarize a conversation over and over, and keeps Moxie's models in step with the Claude Code and Codex apps on your Mac.
- New
Know when Claude Code or Codex has an update. Moxie offers the same models as the Claude Code and Codex apps installed on your Mac. When a newer version of either is out, Home shows a banner with the command that updates it, ready to copy.
- Improved
Model names are easier to read. For example, claude-opus-5-5 now shows as "Claude Opus 5.5" throughout Moxie.
- Improved
OpenAI accounts can use the GPT-6 models, and new OpenAI accounts start on GPT-6.1 Sol.
- Improved
Gemini accounts can use Gemini 3.8 Flash, and new ones start on it. The older Gemini 2.5 models are gone from the list, since Google no longer offers them to new projects.
- Fixed
Claude Code no longer summarizes in a loop. After you used a Codex or OpenRouter account, switching to a Claude account could leave Claude Code believing the conversation had to stay small, so it summarized long conversations again and again. Moxie no longer changes how much Claude Code keeps in a conversation. Long sessions on Codex and OpenRouter still summarize themselves, now when the model says the conversation is full. If a Claude Code chat still does this after the update, close and reopen it once.
Long Claude Code sessions now survive models with smaller context windows, and the first launch got smoother: Moxie helps you install Claude Code and asks only for an email to start the trial.
- New
Long sessions fit smaller models. Claude Code sizes conversations for Claude's huge context window, but many models you route to are smaller. For Codex and OpenRouter accounts, Moxie now tells Claude Code the real context window of the model in use, so long sessions tidy themselves up in time instead of failing.
- New
Moxie helps you install Claude Code. On first launch, if Claude Code isn't on your Mac yet, Moxie shows the install commands to copy and checks that the install worked before moving on.
- New
Remove your license from this Mac. Settings > License can free your license so you can activate it on another Mac. Your accounts and settings stay put.
- Improved
When adding an Ollama account, Moxie explains that Ollama may quietly drop the start of long prompts, and how to raise its context window.
- Improved
When adding a Claude Code account, the setup command can be copied with one click.
- Fixed
A conversation that grew past the model's context window used to stop Claude Code with an error. Whatever the provider, Moxie now passes that error on in words Claude Code understands, so it summarizes the conversation and keeps going on its own.
Moxie now works with OpenRouter, so a single API key gives Claude Code a huge range of models from many different AI companies. You can also choose when usage warnings arrive, and see what's new after every update.
- New
OpenRouter support (experimental). Add an OpenRouter account with just your API key, pick a model, and Claude Code works as usual, thinking and tools included.
- New
When you add an OpenRouter account, Moxie fills in the 20 most popular OpenRouter models that work with Claude Code, so you don't have to look up model names. Press refresh to add more. Models you added yourself are kept.
- New
The usage card for an OpenRouter account shows your key's spending limit and when it resets, plus how much you've spent this month. On a free model, it also shows how many of today's free requests you've used. Your credit balance isn't shown, because OpenRouter doesn't share it with regular API keys.
- New
OpenRouter sometimes asks Moxie to slow down for a moment. Moxie no longer mistakes that for your account running out, so you won't get a false "limit reached" notification.
- New
Refreshing in Settings > Models now shows friendly names for OpenRouter models, like "Anthropic: Claude Sonnet 4.5".
- New
Choose when you're warned. In Settings > Notifications, pick when the "approaching usage limit" notification arrives, anywhere from 50% to 95%. It stays at 90% unless you change it.
- New
Send Feedback. A new button at the bottom of the sidebar takes you to Moxie's feedback board, where you can share ideas and report problems.
- Improved
Settings > Models is easier to scan. Each provider shows its logo, and models sit in a compact two-column grid. Hover over a model to see its full name.
- Fixed
Improvements and fixes with Gemini provider.
- Fixed
Usage limit notifications no longer say "Anthropic" for accounts from other providers.
- Fixed
For an account that reports no usage limits, the usage card now says so plainly, instead of claiming requests are going through.
Moxie 2.0 revamps the app design and streamline functions and settings.
- New
New rich menubar popup UI with usage stats.
- New
Revamped UI with polished components and interactions. Better UX across the app.
- New
Redesigned settings page with categories for ease of use.
- New
New responsive sidebar.
- New
Experimental Gemini provider.
- New
Free trial and licenses. Introduced a free trial and licenses.
- New
Settings > Models lists the models of all your accounts in one place, grouped by provider.
- Improved
Gemini OAuth accounts can load their models while you're still adding them.
- Fixed
Gemini accounts now offer the same models as Google's own model picker, with a separate choice for each reasoning level. Some models used to fail with an error, and Claude models now work correctly.
- Fixed
Fixes latest Claude models not working correctly.
The 1.x releases took Moxie from a simple MVP to a full Mac app. The highlights:
-
More providers. DeepSeek, Gemini (with an API key or your Google account), and Apple's on-device model joined Claude, Claude Code, OpenAI, Codex, and Ollama.
-
Automatic updates. Moxie checks for new versions and updates itself, always asking first.
-
Know where you stand. See your usage limits in the menu bar and on the Dock icon, a 31-day usage trend, estimated API costs, and your subscription plan for Claude and Codex accounts.
-
Always-current model lists. Model lists come straight from each provider, so new models appear without waiting for a Moxie update.
-
Easier account management. Rename, pause, or archive accounts, refresh expired sign-ins in one click, and import existing Claude Code, Codex, and Gemini logins from this Mac.
-
Smooth switching and streaming. Switch accounts or models without interrupting requests already running, and replies now stream word by word instead of arriving all at once.
-
Backup and restore. Save your accounts, settings, and usage history to one file, and bring them back whenever you need to.
-
A more polished Mac app. A redesigned window with a collapsible sidebar, the Playground in a side panel, and a new menu bar that shows each account's usage at a glance.