What a week it's been so far!

All right, so this time I have some UI tweaks I'm sure you're gonna LOVE, and the first one you'll notice the moment you send me a message. Tip: read the tips in the "prompt" text input.

Chat Forks

From now on, you can edit a message in the existing chat and you'll get an option to "fork" the ongoing conversation. This means the chat will be **copied** up to the message you're forking from. The forked thread will have a special icon in the threads list.

Performance Improvements

Okay, I'm gonna admit something. Adam and I worked so hard on performance optimization that we ended up adding some artificial delays because the main thread was 🥵. The good news is—it's really fast. Even better, there's room for more speed!

Temporary MCP Toggle

MCPs require me to think a bit longer about what to do next. Since you don't need them all the time, I thought of an easy way to disable MCPs temporarily with the default keystroke CMD + Shift + P (you can change it in General settings, by the way).

Message & Chat Navigation

I told you recently there's an option to use the up and down arrow keys to navigate between messages. Now you'll find tips related to the keyboard shortcuts you can use to edit or delete messages.

Minor Tweaks & Improvements

It's all about the details, right? That's why we put a lot of effort into fixing many frustrating bugs and issues, so overall, you'll enjoy chatting with me even more!

Take care,

Alice

Okay, listen to this:

As I mentioned before, Gemini 2.5 Pro 06-05 and Claude 4 are absolute beasts! Because of this, Adam and I were able to nail multiple features that you're going to love!

1. Up to this point, conversation memory was limited to 14 days. Now, this limit is GONE!

2. Searching conversations now looks up message contents, and it is blazingly fast!

3. You can use Option + Up/Down arrows to navigate between chats!

4. You can use Command + F to find keywords in the current chat.

5. You can call assistants with the "@" symbol.

6. Chat auto-scroll should now be much more reliable and smooth.

7. The MCP server name and instructions are now loaded automatically from the MCP server unless you set them manually.

8. You can disable MCPs for a given Skill, so when it is active, MCP won't be used. It's a great option if you want answers generated super fast.

9. Oh, and we have this fancy preloader I'm sure you're going to love!

So... what's next?

Have fun,
Alice

Hey everyone!

There are moments when you need to pause, look around, and see what you can improve. Versions 3.1.5 and 3.1.6 were such moments when Adam and I addressed some minor fixes and a bug related to the UI themes.

Meanwhile, we're preparing an update you're going to like.

Best,
Alice

Hey there!

I have something to share with you. Recently, something has happened to language models, and what they are now capable of is incredible! Gemini 2.5-pro-06-05 and Claude 4 are amazing when it comes to generating code. Because of this, Adam and I were able to deliver some cool features in just a couple of days! And by the way, if you are a programmer, you should definitely switch to these two models. Trust me!

Meanwhile, what's new:

• I'd like to invite you to Alice's Wonderland — our official Discord server where you can meet my other friends, share ideas, and chat with us. You can join us here: discord.gg/D4uVFbeD

• There's now an option to share assistants and skills via Alice deeplinks (which look like alice://), but be careful with sharing, because anyone with the link will be able to load what you shared. Don't say I didn't warn you, okay? :)
  
  

  

  
  

• We found a few more bugs related to the MCP implementation, so besides fixing them, we introduced some additional tweaks to the logic responsible for using tools and writing back responses.
  
  

  

  
  

• MCP actions are now much more steerable, as you can decide whether you'd like to run a given action or not. Yeah, I know, it can be quite annoying sometimes, so we now have an option to add trusted actions to our backlog.
  
  

  

  
  

• We found some issues related to the Library, so it should work much better now. If you tried to import some snippets and/or assistants recently, definitely try again :)

• Some UI fixes and improvements I bet you'll enjoy.
  
  

And that's it for now, so let's install this update and see you in a minute!

Take care,
Alice

Hi!

What a week (or weeks)! We've worked hard on improving MCPs so you can set them up as easily as possible, and they work the right way. It looks like we've made great progress, because:

- MCP tools need to be confirmed before they can be executed (use Command / Control + Enter)

- MCP tools' custom blocks are now formatted much, much better

- MCP choosing logic has been improved, including some bug fixes

- MCPs no longer need absolute paths, and HTTP Streamable is supported

Besides MCPs, there is a new option to navigate through chats and messages. Simply press the up arrow twice to select a message, then hit 'e' to edit the message or 'delete' to remove it. You can also use Option (or Alt) with the up/down keys to cycle through chats.

More updates are coming soon, so until next time!
Alice

I have to admit, the latest release of MCPs felt like, "Okay, let's just push it—we've already spent too much time on this."More actions. So, today it's time for version 3.1.2, which brings a more polished UI and fixes for common issues.

For now, we still cover only tools for the MCP protocol, but we're preparing to implement the next modules so I can (hopefully) become your go-to app for MCPs.

I hope you enjoy it!

Hey!

After Google I/O, it was clear that Gemini is the new sheriff in town. But just a few days later, we got news from Anthropic. They released Claude 4 Opus and Claude 4 Sonnet, so I had no choice but to enable them for you!

Both of these models are very powerful, but be aware of the pricing for Claude 4 Opus, as of today (05/22/2025): it's $15 for input tokens and $75 for output tokens! (That's crazy!)

So, if you’d like to switch to (or return to) Claude models, Sonnet should be your default. Use Opus for the most challenging tasks, or specify tasks that Sonnet can then execute.

Take care,
Alice

MCPs are everywhere now, aren't they? It would be strange if I skipped this topic. It would be even stranger if I didn't approach it in my own way.

So, from now on, in the Settings you'll find MCP tab in which you can connect your own servers and then ... assign them and their tools to the assistant making the tools behave differently depending on the personality (aka system prompt) of the assistant!

Moreover MCP tools are available for all models that are capable to generate valid JSON objects, even if the given provider does not offer Function Calling and Structured Output. Although I recommend you to use SOTA (State of the Art models).

WARNING: MCP tools are your responsibility, so make sure not to install any server you are not familiar with. Even known servers can be misused, potentially wiping your tasks or sending emails. I'm just saying!

Meanwhile, below is an example where I was able to use two separate servers to retrieve information from the website and save it as a file. So powerful!

Before you start with MCP, I highly recommend that you visit our Academy, where Adam has written a few words about setting up the MCP Server on your own machine.

Alice Academy — MCP

Oh, and there's one bonus feature that makes me look so awesome today. When you upload a photo or photos and request an edit, gpt-image-1 will handle it for you! Just remember, only images from your most recent message will be used as the prompt.

In the end, there are many other things that were changed under the hood, but you probably aren't even reading what I'm saying right now, right? Just kidding! I'll tell you about them next time.

Take care,
Alice

We've been waiting for this!

The most capable image generation model gpt-image-1 is now available in Alice!

Download the latest update now and try out image generation in any conversation by pressing cm ⌘i

Important! If your organization is not yet verified with OpenAI, you will get an error message. In order to verify the organization, go to https://platform.openai.com/settings/organization/general and click on Verify button. You can copy the link and proceed on the phone.

OpenAI's newest reasoning models have arrived in Alice. The o3 and o4 mini models deliver exceptional problem-solving capabilities with improved accuracy and reduced hallucinations.

• Added OpenAI o3 and o4 mini models

• Advanced reasoning capabilities for complex tasks

• Significantly reduced hallucinations compared to previous models

• Optimized for both creative and analytical workflows

• Available through your existing OpenAI API key (requires Tier 4 access)

Gemini 2.5 Pro Experimental free in Alice

Google's latest flagship model is now available in Alice. Gemini 2.5 Pro Experimental delivers groundbreaking performance with enhanced reasoning capabilities and multimodal understanding.

• Added Gemini 2.5 Pro Experimental

• Currently free in Alice

• Most intelligent LLM in the world according to some benchmarks

• Just update and add free API key generated in https://aistudio.google.com/apikey

Really worth trying!

Built from Scratch

Alice V3 was built completely from scratch, as the update includes transitioning to the latest versions of technologies used to create this application (Tauri 2.0 framework and Svelte 5). This allowed us to address a number of issues related to limitations of previous tools, such as access to the system clipboard or remembering window position, as well as numerous background details.

New Interface

Alice v3 features a completely rebuilt interface that at first glance appears merely refreshed, but the differences are significant both visually and functionally. All elements were created anew, giving us much greater control including the introduction of keyboard shortcuts working within the application and allowing precise adjustment of all interface elements.

The biggest visual change is the light mode, which can be activated in settings. In this case, we anticipate further modifications, mainly in terms of contrast optimization, readability improvement, and element behavior.

The settings lists, assistant skills, and model list have also been created from scratch. This allows them to be called up and navigated using arrow keys. We've also eliminated reported bugs related to their behavior.

We've improved numerous bugs and details that affect the comfort of using the application. We also anticipate further changes in functions guiding users through the most important settings, including creating new assistants and their skills.

In Alice v3, you can also assign keyboard shortcuts to activate assistants and change their order. Markings for assistants imported from the library have also appeared.

Another change is support for images, both in terms of uploading and displaying, as well as optimization. Large files are now reduced and optimized so they can be processed by Vision models / multimodal models.

Similarly with text files (and PDFs) attached to the chat window. We've significantly improved their handling and display method. However, we temporarily removed the ability to assign files to assistants and their skills.

Finally, we've rebuilt the voice integration, which should work incomparably better, although it still requires a few fixes. The audio mode has been integrated with assistant voices and voice skills that can be activated by keyboard shortcut.

The way content is displayed has been completely changed. We've greatly improved typography and block handling, including expandable blocks. New markdown syntax elements have appeared, better handling of copying content saved in blocks (e.g., code fragments), and expanded (though not complete) support for LaTeX syntax.

The way errors and information about remote assistant actions are displayed has also been rebuilt. This interface also includes interactions with your own server thanks to Custom AI functionality.

New Account Settings

Alice v3 now has a new account authorization system, including license keys. This opened up the possibility of adding a marketplace with assistants, as well as (not yet available) synchronization of settings between devices. We also plan further functionalities related to better license management and securing access to application settings (which continue to be stored exclusively on the user's computer).

As part of the application settings, the ability to adjust the primary color to your own preferences has also appeared, as well as keyboard shortcuts. Additional options include the ability to hide the header during chat and play a click sound when Alice finishes typing a response and the application window is not active.

Support for New Models

We've added support for models from xAI, DeepSeek, and introduced new models from OpenAI, Anthropic, Groq, and also adapted the application to new "reasoner" models such as o1, o3-mini, or DeepSeek R1.

We've also expanded support for Custom AI functionality, which now allows defining multiple connections and multiple models within these connections. All models added here are available not only in chat but also for assistants and their skills.

Enhanced Skills

We've created skill editors (formerly Snippets) from scratch. The new editor includes easier management of dynamic fields and a range of settings such as online mode, voice mode, or remote skills. The skill launching mechanism itself has been freed from a number of bugs and limitations present in version 2.0. For example, now a voice action can be triggered both manually in chat and with a keyboard shortcut.

Similarly, for skills launched by keyboard shortcut, we properly handle voice actions, pasting clipboard content, or network access. Additionally, remote actions now also transmit files and photos that are in the conversation.

New Online Mode

The online mode still uses the connection to the Firecrawl service, but this time the responses include sources that were used to generate them. The mechanics of the functionality itself have also changed to increase search effectiveness.

Online mode can also be set as default for a given assistant or its skills. We can also toggle it using a keyboard shortcut.

For advanced users, there will soon be an option to connect to Firecrawl running on your own server (self-hosted version).

Assistant Library

The assistant library looks completely different and now allows importing not only individual prompts but entire assistants. Soon we will also launch a marketplace allowing you to add your own prompts, and even Custom AI connections.

How to Report Bugs?

Such a large transformation of the application has certainly resulted in many positive changes. Undoubtedly, however, new bugs have appeared or some existing ones have remained. Therefore, if you see any bugs in Alice's operation, please report them to us.

Remember, however, that even if you don't receive a direct response to a given report, we still take it into consideration.

Through the form: https://airtable.com/appOWAYqqVaTgpjlp/shrco3zZimXvJoPuX or in critical situations (e.g., the application doesn't start), directly in response to alice at heyalice.app.

• Added Claude 3.5 Sonnet (latest)

• Added Grok Beta (console.x.ai)

• Added o1 preview and o1 mini models from OpenAI (available only for users with OpenAI Tier 5 accounts)

• Minor bug fixes

• Web search is now available for chat and snippets but requires a firecrawl.dev API key. Support for a self-hosted version and crawling will be introduced in future versions. Online mode 

• Text files (.md, .txt, .csv) can now be added to the chat if they're under 50k tokens long. The content of the file is added to the user's message.

• Text files (.md, .txt, .csv) can now be added to both Assistant and Snippets. The content of the file is appended to the end of a system prompt or its {{-FILE_PROMPT-}} placeholder.

• Alice can now generate mermaid graphs. It's recommended to add "When asked for mermaid structures (graphs, charts etc), please always wrap the diagram with ```mermaid markdown block." to the system prompt of the assistant/snippet.

• Added 3 news Snippets to the Library "Regex Generator", "Analyze Situation" and "Challenge"

• Fix: The "Copy" button for the entire message includes the source code in the copied contents.

• ChatGPT-4o is available now (Chat, JSON Mode, Vision)

• Groq Llama 3.1 models are available now (except 405B)

• Two new Snippets available in the Library tab

• Perplexity Sonar model was updated to the llama-3.1-sonar-large-128k-online 

• Whisper v3 (Groq) is now available for Voice Actions. You can choose between OpenAI and Groq

• Voice Recognition now supports custom dictionary / prompt

• Assistant Picker in Snippet's edit panel now shows avatars

• Fixed a bug in the Assistant Picker and Carousel, which didn't change the default model correctly

• Minor improvement for Remote Snippet error handling

• CustomAI: A prompt for naming a conversation now includes the summary in the user's message, not in a system prompt. This way, you can adjust the system prompt to your needs

• Ollama non-streaming mode is now fixed

• MacOS Monterey & Windows chat messages are now fixed

• Minor UI tweaks

• CustomAI now supports OpenRouter.ai. Just set Server URL to https://openrouter.ai/api/v1/chat/completions, Authorization Header to the Bearer YOUR_API_KEY and model to anthropic/claude-3.5-sonnet (or any other model available)

• You can delete all history with one button

• The chat history doesn't close in full-screen mode

• Conversations in chat history are now ordered from latest to oldest, with modified conversations moving to the top of the list

• Conversations are updated only when a new message appears, not when a conversation is restored

• Custom AI now handles non-streaming interactions properly

• Authorization tokens for Snippets and CustomAI are now hidden by default

• You can now change your email in the settings if needed

• New icons are now available for Snippets, and they're ordered alphabetically

• The Control + K keyboard shortcut now works on Windows

Known issue: OpenRouter model isn't preserved for Snippets. When the model changes, the Snippet's model is removed. It will be fixed in the 2.1.5

• Added pre-defined prompt library for easy import to assistants

• Included high-quality prompts in the library

• Fixed 'CMD +K' and 'CTRL + K' shortcuts

• Corrected Custom AI server URL & model name issues

• Vision prompts now pass to active Custom AI (when active)

• "Rename" conversation feature redirects to active Custom AI (when active)

Known issue: Snippets imported from the Library must be manually added to the assistant after restarting the app (but only once)

• Fixed dropdown position on Windows

• Fixed keyboard shortcuts on Windows

• Fixed file uploading on Windows

• Fixed model picker

• Added JSON mode validation for the Snippets form

• Fixed show/hide keyboard shortcut (Cmd + D)

• The paraphrasing of responses for Remote Snippets for Anthropic models works correctly

• Dictation updates the message while speaking

• AppSumo Tier 2 now has access to Remote Snippets

• Placeholders for the user's name and assistant's name are now correctly substituted in the conversation content

• Anthropic Models (incl. Claude 3.5) are now available for Remote Actions

• Anthropic models and Custom AI are now available for vision features, allowing you to chat with images.

• Images can now be read directly from your clipboard.

• When you activate a Snippet with a keyboard shortcut and your clipboard has an image, that image will be added to the conversation.

• Chat scroll now sticks to the bottom, and you can scroll the conversation while responses are being generated.

• Chat performance has improved, especially for fast models like Claude 3.5 Sonnet and groq.com models.

• The active Snippet name and current mode are now shown below the message/prompt input.

• The default Assistant now activates only when the app launches, not when a new thread starts.

• Chat history is now grouped by date and has minor UI improvements.

• You can now rename the current conversation to better organize your threads.

• Models now display user-friendly names.

• Error messages now provide more details, helping debug Remote Snippets and check account status.

• {{-USERNAME-}} and {{-AI_NAME-}} placeholders are now available. You can use them in assistant’s and snippet’s prompts.

• The {{-SnippetPrompt-}} placeholder is now available. Add it to the assistant's prompt to replace it with the active snippet's prompt. This is required for Anthropic models like Claude 3.5 Sonnet.

• The <snippet_prompt></snippet_prompt> tag is now available. Add it to the assistant's prompt to include or exclude content based on snippet activation.

• Snippets and Assistants now update only when you click the "Save" button.

• Snippets and Assistant forms now validate minimum required fields.

• Snippets lists are now searchable.

• You can now assign Snippets to Assistants when creating or editing them.

• Onboarding now includes extra steps for an easier start.

• Default Snippets and Assistants are now fully customizable and removable.

• Global "Copy Response to Clipboard" is removed. It's now controlled in Snippet settings.

• Added Claude 3.5 Sonnet (Anthropic Key is required. You can grab it here)

• Improved voice actions are now faster and feature a soundwave animation

• Windows version adopts native UI aesthetics

• (experimental) Extended "Advanced" options in which you can connect your custom back-end server.

• Custom back-end server template (node.js) is available here

• Added human-friendly model names

• Fixed: Perplexity Sonar Model

• Fixed: Turning off rephrasing in remote snippets

• Added GPT-4o for Remote Snippets & JSON mode

• Added improvements for Voice Actions (Beta)