AI Pals Engine

Amped Horizon

Released: Jan 8, 2026
Platforms
Last updated: 17 Sep 2026, 00:48 GMT

Wishlist on Steam

Watch Set price alert

Updates & announcements

10 posts

11 Aug 2026Developer post
Update V0.10.0: Moving to Runtime V2 + Changelog 🛠️
Hey everyone! 👋 The publictest version has been updated to V0.10.0 . This is a major structural change. Runtime V2 completely replaces the old one. In the past, a lot of Python "glue code" was necessary between AI Pals Engine and the inference engine (llama.cpp). This layer was
Hey everyone! 👋 The publictest version has been updated to V0.10.0. This is a major structural change. Runtime V2 completely replaces the old one. In the past, a lot of Python "glue code" was necessary between AI Pals Engine and the inference engine (llama.cpp). This layer was a necessity then, but it's becoming more and more redundant as the inference engine evolves. The goal of Runtime V2 was to eliminate that glue code entirely. By locking in on the inference engine we simplify things, ensure better long-term compatibility and faster updates.
What this improves for you:
✨Automatic Chat Formats & GPU Layers: Presets will be much less important now. The default preset is set to "auto" mode, so it automatically detects and sets the correct chat format and handles GPU layer allocation on its own.
⚡Better Responsiveness: With less overhead, the app should perform better and be more responsive in most cases.
🛠️New Native Tool Calling Mode: This is now supported for both local and remote text AI models. This means tool calling is determined by the model itself instead of being forced by AI Pals Engine. Keep in mind that not every text AI model supports the new mode, especially older ones weren't trained for tool usage.
Since this changes how the core engine operates, it needs some more testing to ensure everything is still fully supported. Please keep an eye out for any issues or unsupported features and report them on Discord or the Comm Hub so they can be fixed ASAP! 🐛
Once this update is stable, the plan is to further simplify and improve the UI in v0.11.x and hopefully, we can then focus on translating the app in v0.12.0! 🗺️
📝Changelog V0.10.0.0 (publictest Version)
Including smaller updates deployed since the last Devlog:
Engine: Switch to Runtime V2 and generally reduced python code usage to zero
New Feature: Native Tool Call Mode (for local and remote text AI models)
Improvement: Added Model Filter to Add Remote Text Model View
Presets: GPU Layer Auto Mode is now the default
Presets: Chat Format Auto Mode is now the default
Presets: Added Reasoning setting
Presets: Added Additional Stop tokens (removed from general settings)
Chars: Added toolcall-native example
Widgets: Improved tool call visualization
V9.6.0
TTS: Decreased supertonic3 standard speed (helps with cut off phrases)
TTS: Added setting to disable fadein and out
TTS: Added setting to disable multithreading for supertonic3 (tts output skips words bug)
V9.5.0
Assets and Chars: Added Favorites and Filter
Assets and Chars: Added Tags filter
Improvement: TTS ST3 Name includes now chosen language and Char TTS selection box is now sorted by Name
New Feature: Added TTS Text Selection Option (e.g. quoted text only)
V9.4.0 (Current Stable Version)
New Feature: Added full history export in history view (which includes deleted messages)
Improvement: Instruct now resolves dynamictokens (e.g. InstructAction Behavior Node)
Behaviors - Improvement: Made JsonExtractAction more relient to extract json
Behaviors - Improvement: SetDynamicTokenAction is now synchronous instead of async (delay needed otherwise)
Behaviors - Bug: SetDynamicTokenAction not accepting 0 as input
Behaviors - Improvement: ResolveTokenAction can now resolve multiple tokens and text from the text flow
Copy Functionality: Added retry logic to improve reliability in case some other app blocks the clipboard
V<9.4.0
Widgets: Added Conversation Starters to Widget Sidepanel + the option to hide it
Widgets: Math Latex Renderer (Optional: Enable it via Widget Config -> Use Syntax Highlighting and Math Latex Renderer in Text Highlighting)
Widgets: Improved Rendering and Output Panel Scrollbehavior
Improvement: Presets now support KV Cache Quant Types (Context Scaling Tab)
Improvement: The Test TTS View now breaks sentences like the normal Text Generation when generating the Audio
Improvement: Pauses after each text segment split before continuing playback can now be configured for TTS
Char Config: New Tokens -> current_date, current_time, current_datetime, current_weekday, current_weekday_short (see token overview)
Bug: Export as PNG, fixed error on empty conversation starter
Bug: Inspect Context Help Button not working
Bug: Chatview/Testview -> `Remove Last Message` after Stop sometimes not working correctly
Bug: The Char Config `Set Image` now works correctly in case there are pending changes
Bug: Char Overview now properly displays the latest Char preview Image after Char Config `Set Image`
Bug: Chat Format wizard suggesting wrong Qwen3.5 format
Bug: GroupChat AutoMode fails to determine participants after switching Char
Bug: Fixed GetNoteAction not working
Bug: chat-rp-morganlefay replaced wrong {{char}} with {{user}}
Bug: If the history or option count is zero the history sidepanel section will now hide properly
Help: Added Experimental Page
Original post

03 Apr 2026Developer post

Google Gemma 4 & Devlog #5: State/Mic Check, Workshop and Bundles + Changelog

Happy Easter in advance 🐰! Yesterday, Google gifted us Gemma 4 in multiple variants and experimental support has been added to the Steam publictest version and other editions since then (see inapp Recommended Models ). Model Parameter Size Context Window Gemma 4 E2B 2.3B effecti

Happy Easter in advance 🐰!

Yesterday, Google gifted us Gemma 4 in multiple variants and experimental support has been added to the Steam publictest version and other editions since then (see inapp Recommended Models).

Model	Parameter Size	Context Window
Gemma 4 E2B	2.3B effective (5.1B w/ embeddings)	128K
Gemma 4 E4B	4.5B effective (8B w/ embeddings)	128K
Gemma 4 31B	31B dense model	256K
Gemma 4 26B A4B	Mixture-of-experts (4B active / 26B total)	256K

Since the Early Access release, a lot has changed and many improvements have been made. Now is a good time to take a step back and look at where things stand.

Are there any features previously mentioned that you feel are still missing (aside from those outlined in What’s on the Horizon for AI Pals Engine)? Or are there bugs you’re still encountering? Your feedback is incredibly valuable in helping shape the next steps.

Hopefully, you’ve already noticed the progress across this and other editions of the app... and that your feedback is making a real difference!

For this devlog, I want to focus on a new feature that has been requested since day one: Voice Activity Mode, as well as take a look at the current state of the Workshop and the Bundle concept.

An Image Pal Bundle Test (available to you via the Workshop). The images are randomly replaced via Behavior (SetPalAvatarByTagAction using tag AImage) to achieve e.g. a different Pal per Session.

🎙️Voice Activity Mode

Using the microphone without needing to press a button has been one of the most common requests.

I generally agree with the Discord that multimodal LLMs with fully integrated Audio are the future and hopefully the multimodal trend continues. Gemma 4's addition of native audio support (the small E models) is another reason to experiment further with implementing native audio support.

But until we fully get there, I wanted to improve the current mic experience with something that will remain essential even in a multimodal future:

Automatic speech detection

In other words, this is a mic auto mode. Once Voice Activity Mode is enabled, you can simply start speaking - no button presses required - and the system handles the rest.

How it works

In its current implementation, a lightweight model continuously analyzes the audio input to detect when someone is speaking.

Detected speech is buffered and once you stop talking, it gets sent to the Speech-to-Text model. From there, the result is passed to the AI model and the response is returned as text or TTS.

There is still some noticeable delay, but especially with CUDA enabled and a performant model, it's already quite responsive. At the moment, a few basic voice commands are supported, along with an optional wake word.

Let me know how it performs in your testing!

🛠️Workshop & Bundles

You already have access to it as it evolves, but the Workshop isn't officially released yet. There is also an EULA related bug that blocked new submissions, but that has now been fixed in the publictest version.

The main reason for holding back the full release is simple: I want to spend a bit more time refining the system and to understand which tags are actually useful.

Recently, improvements have been made to the UI (Current support: Pal Images and Chars). I'd love your input here: which asset category should be supported next?

Bundles

Bundles are essentially collections of related Chars and Assets. For example:

The same Character across different scenarios
A Pal Image set with multiple emotions or visual variations
Specialized assets like stylized button image sets for action buttons

Bundles can also be shared via the Workshop, making it easier to distribute sets of content.

As shown in the initial GIF, there are also specialized behavior nodes that allow you to:

Retrieve a random Char or Asset from a bundle
Filter results based on tags

For example, you can dynamically switch expressions based on emotion (see Widget Example Advanced 1 – Lightweight Expression) or load a different Character scenario each time a widget is reset.

This opens up a lot of interesting possibilities. Imagine pulling in external data, like retrieving the current weather and automatically selecting a matching Pal Image.

Looking ahead

The Workshop has the potential to go far beyond just Char and current asset types. In the future, it could expand to include:

Full Widgets
Themes
Small, game-like widgets
Abilities
Tools

I'm especially interested in experimenting with more stateful, game-inspired widgets, systems that follow rulesets, similar to tabletop experiences like DnD. It would be wonderful if, in the future, we could share even small minigames. First things first, though: let's start by integrating more standard assets.

🌅What's on the Horizon for AI Pals Engine?

Right now the backlog looks like this (excluding personal things I want):

Chatformat Auto Mode and Overhaul (Long Term)
Long Context Performance Improvements (+KV Quant support)
More of the following: Help pages, Internationalization, Fixes, QoL Improvements

📝Changelog V0.9.1.1 (publictest Version)

Summarized from smaller updates deployed since the last Devlog:

✨ New Features

Plugins: Added Voice Activity Mode and Editable Voice Commands
Gemma 4 Support (V0.9.1.1)

🔧 Improvements

Compatibility of the different AI Pals Engine Editions
System Tray: Changed the style to a Dark Theme
TTS: Added subtle fadein/fadeout
Memory: MD File Support (V0.9.1.0)
Settings - Audio: Added TTS Volume Slider
Settings - Audio: Added Widget Volume Slider
Settings - Audio: Added Microphone Id
Widget: Choosing a Char or Text Model now autocloses the sidepanel (+ added key shortcuts)
Chars and Assets: Added Bundle Filter
Workshop: Added Search Text
Workshop: Added Favorite Button
Workshop: Added Pagination
Workshop: Improved and fixed Sort and Filter Options
Help: Added Multimodal Model/Plugins Help, modified Get Started and improved inapp tooltips

🐞 Fixes

Workshop Submit stuck on EULA check (V0.9.1.1)
Widget Dialog Cancel+Ok button are now more consistent
Widget Alt+F4 now properly Unsets the Active Space

Original post

15 Mar 2026Developer post
Devlog #4: AI Pals That Remember You + Changelog
Happy Sunday! ☀️ The latest update introduced the automatic mode for GPU layers . This is also planned for the chat format , which should make things a lot easier (as promised in the Early Access description). If you run into issues with automatic mode, let me know, it will likel
Happy Sunday! ☀️
The latest update introduced the automatic mode for GPU layers. This is also planned for the chat format, which should make things a lot easier (as promised in the Early Access description).
If you run into issues with automatic mode, let me know, it will likely become the default in later versions. Also I'm curious: How are the new Qwen3.5 models performing in non-English languages? I was positively surprised how they performed in my image interpretation tests.
This devlog is about the memory functionality, which is a bit of a complex topic and the access to it a little hidden, so let's look at how it works.
🧠AI That Remembers
Imagine talking to an AI Pal that remembers details about you. Not just within a single conversation, but across long chats, roleplays or documents. That’s the idea behind the memory system.
The Context Window Problem
If you've used AI and AI Pals Engine, you know the context window: the maximum amount of text a model can process at once.
A few years ago, getting 4096 tokens in a local model was already impressive. Today we have much larger contexts, but they still have limits and roleplay world descriptions, documents and long conversations can exceed them (and of course your hardware).
Instead of trying to fit everything into the context, the memory functionality uses a different approach.
Memory as Search
The system works similarly to the web search tool we look at in this Devlog, but instead of searching the internet, it searches a local memory database.
Example
User question:
"What was the name of the city my character came from again?"
Due to semantic search the system can retrieve relevant entries, even if the original text looked like:
"Elira was born in the coastal city of Vaelthorne."
Why This Matters
The AI Model itself is static and can't add information on the fly. Using the memory approach allows the AI to access information that goes far beyond the model's context window. Without memory systems, long-running interactions eventually lose important details.
Practical benefits:
Conversations stay coherent over time – even very long chats won't lose key information or earlier decisions.
World lore and documents stay accessible – roleplay settings, guides, manuals or research notes can all be referenced during chat.
Remembers important details – characters, story events, personal preferences or ongoing projects can come up again later.
Large documents become usable in chat – instead of pasting them repeatedly, the AI can retrieve relevant sections when needed.
Your own knowledge base – you can build a personal database of information the AI can search and use as a general reference.
Less repetition – you don’t have to keep re-explaining the same context or background information.
✅ Can I Use Memory in AI Pals Engine?
Yes, you can enable Memory in the Char's Config. Your input will then be compared with entries in the database using embedding similarity.
At the moment:
Each Widget uses a shared memory (Char-independent)
The chatview/testview has memory per Char
You can access the memory view in the sidepanel, where you can:
Paste in text
Load documents
Manipulate current memory entries
You can also store conversations that are removed when the context becomes too long, this is configurable in the Char's Config.
When memory is enabled, the {{memory}} token is automatically replaced with actual memory entries.
Example
This Char's memory includes the entire American Bible Standard Edition, which is a little less than one million tokens of text.
For the test we'll add additionally: the secret word is 'Curiosity'
👉 Out of 5578 text chunks, it still is able to retrieve the right information to phrase the correct answer to "What is the secret?"
🔍 How Could We Further Improve the Memory Functionality?
The current system uses similarity-based comparison, which means you might not always get the most relevant entries, just like how web search doesn't always return perfect results. So, there's always room to improve lookup accuracy.
Even more importantly: it would be powerful to build functionality on top of memory entries. Imagine an AI that:
Continuously transforms new memory entries into meaningful insights
Compresses information so you can inject more (or all) information
Learns about your RP history or personal patterns
Analyzes your behavior
Right now, you can experiment with this manually:
Multi-select messages → copy → transform via AI → re-enter into memory
Enable “Auto-load the last conversation” in the widget settings to ensure the conversation grows long enough for messages to be transferred to memory.
❗As with any of the features, Memory will become more integrated over time and easier to use. Let me know if you think it's worth adding an automatic layer, maybe using a fast, lightweight model to handle tasks like summarization or compression automatically!
🌅What's on the Horizon for AI Pals Engine?
Speech-To-Text Wake Word
Better Audio integration
More of the following: Help pages, Internationalization, Fixes, QoL Improvements
📝Changelog V0.8.18.0
Summarized from smaller updates deployed since the last Post:
✨ New Features
Presets - GPU Auto Mode: Auto GPU Layer Selection + MoE Speed Optimizations (when enabled)
Presets: Basic Device Support (GPU selection)
Extended Model Support: Added Qwen35 and other hybrid model support
Char Config: Available Dynamic Tokens Overview
Char Config: ST Card Exporter/Importer
🔧 Improvements
Char Config: Increased text limits
Char Config: Aiprepend/Userappend now parse dynamic tokens
Char Overview: "Load in Widget" now temp-loads and starts a fresh conversation (instead of just selecting it in the widget settings)
TTS: Better handling of non-English-only sentences
Memory: Multiselect + enhanced info & usability
Help: Group Chat and the Memory pages have been improved
🐞 Fixes
Fixed Widget Temporary Char dependency loading
Fixed duplicate key error on context overflow and chat memory enabled
Original post
25 Feb 2026Developer post
Model News + Changelog: Qwen 3.5
AI Pals Engine allows you to hook in any model through a custom endpoint, but for convenience, it aims to support as many local models as possible natively . If you notice a model isn’t working smoothly or a new model release isn’t getting the spotlight it deserves, head over to
AI Pals Engine allows you to hook in any model through a custom endpoint, but for convenience, it aims to support as many local models as possible natively. If you notice a model isn’t working smoothly or a new model release isn’t getting the spotlight it deserves, head over to the Community Hub to let us know. Please also let me know if you would like to receive more information about model releases via news section or in-app notification.
The Qwen Team has dropped some exciting new models:
Qwen/Qwen3.5-122B-A10B
Qwen/Qwen3.5-35B-A3B (Yes, MoEs aren’t everyone’s cup of tea, but I’ve been using the predecessor heavily for my tasks. It’s fast, efficient and plays well with systems packed with RAM)
Qwen/Qwen3.5-27B
Smaller variants are coming soon too.
These models are built on a new architecture featuring:
Improved long-context handling and more efficient attention mechanisms (=improvements/faster for lots of text)
Native multimodal support (text + vision)
Expanded language coverage: 201 languages and dialects
Due to the newer architecture, these models still require some runtime improvements. If you'd like to test them early (this also applies to other Mamba/Hybrid models), please switch to the publictest branch (Properties -> Game Versions & Betas) in the Steam menu (V0.8.14.0). Several runtime updates will be made to this branch (in addition to other improvements) before it is stable enough to be included in the normal release.🛠️
📝 Changelog V0.8.13.2
Summarized from smaller updates deployed since the last Devlog:
General: Added --nosteam argument to the executable
TTS: Added adjustable Speed/Rate settings
TTS: Updated the chatterbox project to support the multilingual model
New Tool: screenshot_dx11 (note: the previous screenshot tool may fail in certain edge cases)
Widgets: Improved Markdown rendering during streaming (fewer artifacts)
History: Optimized loading for faster performance
Add Text Model Endpoint/Server: Added test button for easier debugging
Globalization: Added language switching functionality and selector test
Improvement: Chatview now uses always up to date textmodel info without reselection
Fix: "Continue Chat" now works reliably on the first try
Fix: Default TTS model no longer loads unnecessarily when TTS is enabled
Fix: Retrieve Text Model List Button (and TTS) only using the "dummy" key
Fix: Multilingual TTS Text Encoding Bug
Fix: Broken Add Text Model Save Button
🌅 Here are some points for the next updates:
Basic Multi-Gpu config
Better Memory+Dynamic Tokens integrations
STT Wake Word
Better Audio integrations
More of the following: Help pages, Internationalization, Fixes, QoL Improvements
Original post
08 Feb 2026Developer post
Devlog #3: Changelog + Letting AI Push Buttons with Tools (What Could Go Wrong?)
Happy Sunday! You might recall we used an MCP Tool to generate images in the last Devlog about Behaviors . This Devlog takes a closer look at how that actually works. Tools/Agents/Skills are a hot topic right now. Frontier models have improved drastically in recent months, partic
Happy Sunday! You might recall we used an MCP Tool to generate images in the last Devlog about Behaviors. This Devlog takes a closer look at how that actually works.
Tools/Agents/Skills are a hot topic right now. Frontier models have improved drastically in recent months, particularly regarding coding tasks. But most of these advancements can also be used in a more general way to do all kinds of things. We're getting dangerously close to the point where I can replace myself with a coding agent and simply supervise it! Anyway, we will see this expand into more capable open models in the future as more training data gets integrated for this task, in addition to native capabilities within the chat format.
A tool call to retrieve realtime information from the web
🛠️What is Tool Calling?
Ultimately, it is just a program following a protocol (AI Pals Engine uses MCP - Model Context Protocol): the Text Model generates the command and the tool executes it.
Imagine a tool that lets the AI execute Windows commands directly. Is it smart to give your AI that kind of power? Probably not.
✅Can I do Tool Calling in AI Pals Engine?
Yes, support is much broader after the recent update and hopefully you can add any tool. But as mentioned within the app, always use only trusted tools. They carry a lot of risk!
You can opt for a Single-Step approach (one tool call) or a Multi-Step workflow where the AI executes a chain of calls to complete a task.
You also have full control over how the data is handled: whether the result is added to the context, used immediately in a response, or both. There are also different viewing modes: Raw (to see every detail) or Compact (to save tokens and keep the format pretty).
Beyond the basics, you can add tools into group chats to create teams of "Agent-like" characters. There are also special tools like send_message_to_user (for status reports) and end_task (to stop chains). To make sure everything runs smoothly, AI Pals Engine uses grammar enforcement by default, guiding the model to ensure it always produces valid commands.
Example 1 – Realtime Info & Grounding
Large Language Models are famously "frozen in time", they only know what they learned during training. If you ask them about news from yesterday, they might hallucinate or give you outdated info.
This is where Grounding comes in. Grounding essentially means anchoring the AI's response in reality by providing it with external, up-to-date evidence. You can do this via the Memory Feature (providing text documents), by using a real-time search tool like the example above (the example utilizes SearXNG) or by manually providing the info. Instead of using the send_message_to_user tool, we changed the mode of the search tool to "ActOnResult" to compile a summarized answer using the retrieved information.
Example 2 – Let your AI do some work by chaining tool calls
Why browse the web yourself when your AI Pal can do it for you?
By utilizing Playwright (a general browser automation tool), the AI can navigate the web, click buttons and interact with pages. The magic here is the Chaining capability. The AI doesn't just fire one command and quit, it intelligently strings together a sequence of actions (browsing, clicking and reading) to complete complex tasks entirely on its own. But don't get too excited, a small model will fail to do things (if it wasn't trained especially for it) and the context usage can be really high for this task (a lot of computation).
Other Examples
Like with Behaviors, Tool Calling allows for endless possibilities to do more than just text generation with your Char. As mentioned in the intro, these capabilities will only improve in the near future.
I personally use it with Github Tools and to enter or modify Notes/TODOs. I plan to experiment much further - for example, generating predefined choices that get automatically forwarded to a widget panel. We could even implement a standard for interpreting returned results to trigger other panels or display info. This would better support both game-like and general interactive functionality, such as tracking game states, inventory or skill stats, system states, user data and other dynamic information displays.
Dialogue Options
🌱Future Plans
This thread is about how to make Tools/Skills discoverable. For Steam, the most convenient way would probably be the Workshop, so that we can just pick the Skills we want in the Char Config. But the main issue is: how to make it secure? We want the tools to be powerful, but at the same time, they can be used to harm your system in many different ways.
For now, the best approach is likely to make the tools of the AI Char shareable to the Workshop with metadata on which tools were used and where to find them.
Another idea is to add a UI panel to temporarily disable/enable tools (e.g. the web search tool). Sometimes you need it, sometimes you don't and it is probably faster to use a panel than to switch the AI Char entirely (of course you can leave that up to the AI, but I usually prefer to have the control over it).
Let’s keep an eye on advances and reassess in a few months, once more models are available on newer standards.
🌅What's next for AI Pals Engine?
Hopefully, this Devlog gave you a better idea about the Tool concept in AI Pals Engine. You can also use tools manually in your Behavior workflows. For the Text AI, there is just a basic tool call example available within the app (chat-general-toolcall), but with some tweaks and by adding tools, you can make a really powerful Widget!
The question remains: Would a skill workshop be a good addition? Also, what about security? You can comment here or use the mentioned thread.

Here are some points for the next updates:
TTS SPEED Config
STT Wake Word
Basic Multi-Gpu config
NoSteam Option (disable Steam Time Tracking) and Autostart
Better Audio and Memory+Dynamic Tokens integrations
More of the following: Help pages, Internationalization, Fixes, QoL Improvements
📝Changelog (summarized from last to V0.8.11.0):
TTS: Improved Piper TTS support (load any model/language).
TTS: Fixed Kokoro not working without eSpeak.
TTS: Fixed default Kokoro not using the correct text-split strategy.
TTS: Optimizations and better handling of sentences including the "-" character.
Widgets: Added option to automatically load the last conversation on startup.
MCP: Reworked "Add MCP Server" to improve usability.
MCP: Added support for Environment Variables (needed for custom Python env servers).
MCP: Added basic HTTP support.
MCP: Added support to dynamically generate the method for the McpAction Behavior Node.
Tools: Added OAI Image tool.
Tools: Added Web URL POST tool.
Tools: Reworked Character Tool Config to improve usability.
Tools: Fixed send_message_to_user not outputting correctly in Raw Tool Call Mode.
Workshop: Character Tool allow-lists are now exported (Tool Call Mode defaults to "None" for security).
Workshop: Fixed Workshop items not being deletable if removed manually from Steam.
General: Added more help text and tooltips.
General: Added warning when exiting the app if a download is active.
Performance: Fixed performance issues when loading many entries in the Memory View.
Original post
18 Jan 2026Developer post
Devlog #2: From Behaviors to Meows + the Changelog
Happy Sunday! Behaviors are one of the things I'm most excited about. They form the backbone of a truly dynamic Pal - enabling automation, unlocking deep modding possibilities, serving as a strong starting point for custom functionality and further enrich our Widgets . That's why
Happy Sunday! Behaviors are one of the things I'm most excited about. They form the backbone of a truly dynamic Pal - enabling automation, unlocking deep modding possibilities, serving as a strong starting point for custom functionality and further enrich our Widgets. That's why I'd like to dive into them a bit more deeply, with real examples. At the bottom, you'll also find the changelog, detailing the fixes from the recent smaller updates.
Behaviors are essential to the core, every update on them opens new doors for Pals and Widgets. The real magic shines when you take them further: crafting custom behavior, testing limits and adding/creating entirely new things.
Example - A Widget Behavior that initiates the generation of an image (external tool) to visualize the current scene:
✨ Moving beyond Language Models in AI Pals Engine
Language models are fantastic: they handle language processing, can mimic personality, keep conversations flowing and understand high‑level intent. But we need
While language models act as the soul or brain of a Pal, we also need a "nervous system" that executes actions like animations, movements, sounds and more. Not only does this system power the Pal, but it also enriches our Widgets with more interactivity and functionality.
Combining language models with Behaviors gets us closer to a true mind‑and‑body avatar. Behaviors serve as the nervous system: fast, predictable and rule‑based. They’re built on Behavior Trees, a widely used AI decision-making method in gaming (More AI on top of AI in AI Pals).
🧩 Are they available in AI Pals Engine?
Yes, you can edit them, create them and share them. You'll find them by clicking on Behaviors in your Pal's Avatar Config within the Widget Config (an avatar has to be set). You can even reuse your existing MCP tools as mechanisms to trigger actions on other systems.
And yes, Behavior Trees aren’t exactly intuitive at first, especially given the current basic implementation and editor. But they're a well-established foundation. We can later add an ability-like system where we simply pick pre-built trees. And AI can help build them or even better: what if the AI modifies its own “nervous system”, dynamically generating or evolving Behavior Trees on the fly?
Let’s cook some more on it. Anyway, enough theory let’s get into some examples.
Pal Example
This is my favorite Pal (Example Widget Name: No AI 2 – Cat), it makes me happy, you can pet it and it plays a sound via Behaviors 🐱:
What if we want it to meow randomly? Easy we just replace the Interaction-Trigger Node with a Random-Delay Node:

I use it personally as a notification system. I get 10 meows for every review and 1 meow for a new entry in the comm hub. How?
A custom-coded MCP Tool that checks both counts - combined with this Behavior (+ a Repeat Node set to 10 repeats to trigger 10 meows).
Widget Example
As mentioned Behaviors can also serve to improve the Widgets. Let's take a look at the Advanced 5 – Dice Adventure example. Here, we use Behaviors to inject dice rolls and introduce real randomness into the AI model’s decision-making. For instance, if we encounter a monster and roll a negative outcome, your character could die right then and there (Make sure not to use a completely positive biased model).
In the future, I plan to do more experiments around this and more complex game rules. Maybe introducing state systems to manage story arcs, inventory and abilities, all of which could influence the randomization and deepen the narrative experience.
Image Example
So, how can we recreate the example from the beginning of the post? It follows the exact same logic and is pretty easy, the overall steps are:
Determine when we want to get the image. After the AI completes a message or do you prefer triggering it manually via an action button? (Use the GetLastEventDataAction node)
Retrieve the image via a custom MCP Tool (which also enables Pals to generate images) or through an endpoint using an McpAction node (AIP-MCP.get_from_url_image)
Set the new image as the avatar of a decoration Pal or your current Pal (use the SetPalAvatarAction node).
The tricky part is integrating a local running image gen functionality. The plan is to add a tutorial to the Behavior help section soon - making it super-easy to recreate this on your own hardware, without the usual hassles and dependency headaches (or in Todd's words: it just works) + the importable Behavior.
🌅What's next for AI Pals Engine?
I hope I was able to communicate the power of these concepts. We'll definitely need some quality-of-life and usability improvements. If you want to experiment, go ahead, particularly let me know if any crucial nodes feel missing!
Release means a lot of troubleshooting that costs a lot of time and are hard to predict. Sometimes AI Pals Engine fails, but sometimes even Windows fails with corrupt .NET environments. So slowly we can start on new things again. Here are some points for the next time, shifting to the things we are discussing in the Comm Hub/Discord, like the following:
Better MCP Config and more Char synergy + Workshop support
Multilingual TTS
TTS SPEEEED Config
STT Wake Word
Basic Multi-Gpu config
NoSteam Option (disable Steam Time Tracking)
Autostart AI Pals Engine and an option to pick up the last conversation automatically in the widget
Better Audio and Memory integrations
More of the following: Help pages, Internationalization, Fixes, QoL Improvements
Let me know what you need and if you’ve had the chance to experiment with Behaviors yet!
Changelog V0.8.7.9
Summarized from smaller updates deployed over the week:
appLog.txt is now more verbose in case of unhandled exceptions and a proper error message got added
Debug Window: It is now less verbose by default (You can reenable it in Settings -> Chat)
Many UI Naming and Usability improvements like:
Avatar renamed to Pal Avatar in Widget Config, Added Copy Text Button in Debug Window, Proper Warnings for Widget Foreground and Mouse Interaction Settings, Chars -> Added Details to context menu, Preset -> better distinction of set vs unset model and added button to reset the default preset, better clipboard error handling, additional messages to explain the app's functionality
FIX STT: Button not changing the color properly when in Record mode (main app)
FIX Char Details: A case in which Pending Changes were automatically displayed when entering the details without changing anything
Fix: Char Name synchronization between views
Original post
08 Jan 2026Developer post
🎉AI Pals Engine is available now!
Time to celebrate! Explore all the features and join the fun experimenting with technology. The FAQ is live with extra details to help set your expectations. Take a look before you jump in and let the adventure begin! And don't forget to share your thoughts and feature requests i
23 Dec 2025Developer post
Official Release Date
Hey all, one last thing for this year and an addition to the last Devlog ! We finally have a release date for AI Pals Engine on Steam: January, 8th! Of course 2026 , I promise! If you don't have it already, remember to wishlist to get notified! Wishing you all Happy Holidays! Tha
19 Dec 2025Developer post
Devlog #1: Widgets in AI Pals Engine
Hey everyone! In Devlog #0 , we discussed why AI Pals Engine was created. Now, in this follow-up, we're diving into one of its core features: Desktop Widgets. Discover what they are, explore the types you can use, and see how they'll transform your PC into a truly personal space
Hey everyone! In Devlog #0, we discussed why AI Pals Engine was created. Now, in this follow-up, we're diving into one of its core features: Desktop Widgets.
Discover what they are, explore the types you can use, and see how they'll transform your PC into a truly personal space - interactive and uniquely yours. Let’s get into it!
Bring Your Desktop to Life
🌟 What Are Widgets?
You know them as small, handy tools that sit right on your desktop, giving you quick access to what matters to you. Instant info, right where you need it.
In AI Pals Engine, widgets are not just simple panels - they’re dynamic containers for customizable content panels, letting you pick exactly what you want to see and use. All organized within a switchable Widget Space.
With AI Pals Engine Widgets, you can:
✔️ Place them anywhere on your screen
✔️ Fully customize their look, function and behavior
✔️ Use them as floating overlays (Foreground Mode) with or without Mouse Interaction
They’re your personal desktop extensions - flexible and built to fit your needs.
🛠️ Types of Widgets
Widgets are designed around your AI Pal Avatar - whether you're going lightweight (pure web tech based), 2D (using Phaser) or full 3D (Three.js). Or maybe you don't want a Pal Avatar at all! The choice is yours. Plus, there's a Webpage Render Widget - great for embedding external content, like dashboards or live feeds. And with shortcut support, you can toggle them on and off instantly.
No bloat. No forced features. Just plug in the widgets that fit your vibe or experiment with the pre-built examples. And let's add some cool new Widgets during the Early Access!
🎯 The Heart: Panels & Pals
While Widgets are a type of container, Panels and Pals are the dynamic components that bring them to life - custom building blocks you place directly into your widgets.
Pals
Your AI Pal, OPTIONALLY powered by a text model and/or represented by a visual avatar like images, shaders, spritesheets, videos or 3D models - all static or behavior-driven (more on that in a future devlog).
Panels
These are the functional building blocks - everything not a Pal. Here's what you can use:
Text Panels: Go beyond static text. Display time, date, CPU usage... or dynamic content using dynamic tokens (explained in a future devlog).
Action Buttons: Click them to trigger actions
App Actions: Minimize the widget, trigger the sidepanel and more... bring frequently used features out front.
Manual Tool Actions: Run specialized tasks on your machine or externally (just like your Pal can trigger them).
Transformation Actions: Perform quick, non-chat tasks with one click - proofread, format, summarize or manipulate text in any way.
Note Panel: Your flexible workspace for long-form content. Supports Markdown/HTML, perfect for notes, todos or live info display.
AI Panels: Input & output panels for your text AI, with multiple variants to match your needs.
With the panels, you can use your Desktop in a new way and also have the flexibility to create Interchangeable Environments (e.g. different roleplay spaces).

It's a reimagining of what your PC can be. The more you explore the App and get comfortable with its features, the more amazing things you can create. Don't want to dive too deep? No problem, there are ready made example widgets.
Turn your screen into a living space: Display what matters most to you... real-time stats, AI-generated text, personal notes or even live feeds.
Add your own workflows: Time, buttons, AI inputs and outputs... all can trigger full sequences of actions.
Blend style and function: Make your desktop reflect your mood and vibe. Switch to a work-focused widget space when you need to focus or a roleplay setup when it's time to have fun.
Share and grow: Export, import and share widgets with others. Discover new setups, inspire others, or create something so unique it becomes a standard.
🔍 Deeper Dive
We explored the WHY in the first devlog: AI Pals Engine is designed as a highly customizable platform for diverse tasks. The powerful flexibility of Widgets and Behaviors makes this possible. However, it's important to remember that AI Pals Engine isn't a pre-built experience tailored to a single purpose; instead, it's a sandbox where you select and integrate the components you need.
To leverage more complex features, some investment in learning the underlying concepts is required. For example, the included RSS example demonstrates functionality, but understanding how Behaviors operate is essential for integrating it into your own spaces.
Currently, Themes and Styling aren't deeply integrated with all elements, such as the General Text panel. Styling often requires direct CSS/Stylesheet manipulation, though it is planned to improve this over time, potentially incorporating AI-driven UI generation.
Regarding performance: Do you need a high-end PC? It depends on how you intend to use it! Simple AI pals that are used purely for decoration can run on a Potato. The standard AI Model that is included will also do its job in a reasonable timeframe, even on toaster to mid-range setups. However, utilizing the latest, state-of-the-art large AI models will demand a rig that is beyond your setup. A mid-range AI setup with 3D Pals should perform well on a high-end gaming PC. Fortunately, AI Pals Engine offers energy management and optimization options. For instance, widgets default to pausing after 14 seconds of inactivity and you can minimize them when they aren't needed.
AI Pals Engine handles all the underlying processes, including loading AI models (even simultaneously), text-to-speech, rendering, Behaviors, chat history and more. This complexity is managed internally and if you find any features or options missing, let me know - that's the benefit of the Early Access Release!
🚀 The Future
What kind of Widgets would you like to see? There's so much room to get wild and use the EA to build new Widgets alone. Can't wait to share more with you! In the end, it's not just about the visuals. The behavior matters too and that deserves its own Devlog. If you're interested in participating a test before the EA Release, please don't hesitate to send me a DM on Discord (or contact me directly somewhere else) for more details.
Thanks for reading and supporting this journey. Have an awesome weekend!
Original post
07 Nov 2025Developer post
Devlog #0: Why AI Pals Engine Was Created
Hey everyone! While the Steam description page gives a brief overview, devlogs offer a deeper dive into the thinking behind AI Pals Engine. This is also a space for you to share your thoughts and feedback along the way and the different features! AI Pals Engine allows you to tran
Hey everyone! While the Steam description page gives a brief overview, devlogs offer a deeper dive into the thinking behind AI Pals Engine. This is also a space for you to share your thoughts and feedback along the way and the different features!
AI Pals Engine allows you to transform your desktop into an interactive, customizable living space, complete with optional AI-powered Pals that are truly yours.
❓The Why
As someone who uses AI daily, I craved a single, customizable app for everything: AI interaction, UI elements that fit the current task, the ability to use ANY AI model (also in combination), advanced features like TTS, interactivity and most importantly, it had to store everything on my machine.
But it goes further. Here are some questions that spurred the creation of AI Pals Engine:
With multi-screen setups, can't we make the desktop itself more useful?
Can we make AI more interesting by giving it a graphical representation and more personality?
Can we leverage widgets to display and share information with AI?
Can we make it easily switchable to fit a specific mood or task?
Language models are powerful, but can't we enhance them with layers proven in game development to create smarter, more lifelike AI?
Can several Pals work together and can each of them use different settings/models?
Isn't the Steam Workshop better than "BUY THE LUNA CHARACTER DLC”, “BUY THE LUNA VOICE PACK DLC”, …
Also, I’ve never liked the typical corpo product cycle – it’s what killed many of my favorite game franchises. Most AI products will also follow the same pattern and feature a “milking” model: subscription-based, locking you into their servers, rules, and data collection practices. Cloud AI is great - until it isn’t: Until your favorite AI model disappears. Until your favorite model is quietly dumbed down to save costs while you’re still paying the same. Until your data gets sold or used against you—learning your habits, personality, and weaknesses for marketing. Until the servers go down or maintenance makes the service inaccessible. Until the “free” version is just a funnel to upsell you. And Open-source? Awesome, use it! This is the way! But keep an eye on security and avoid unsafe projects. With AI Pals Engine you get a safe, easy and quick way to get started + it bundles a lot of functionality in ONE app, as an expert you will be able to do even more.
AI needs the indie scene like games need the indie scene; otherwise, the future is like this: You ask, “What should I do about x?” The answer: a sponsored pitch, not a solution. If it’s free, you’re the product—and AI will turn that into the ultimate targeted sales machine. Anyway, custom endpoints are supported and I’m open to adding more convenient ways of using powerful cloud models (AS OPTION), I just want to make clear why AI Pals Engine will be always offline first.
Lastly, many people using powerful cloud-based AI models aren't aware that their own machines can achieve similar results. Understanding how AI works improves their output and AI Pals Engine is a tool that makes local AI accessible while providing guidance along the way. Hopefully, it will help to increase the adoption of local AI and improve fundamental understanding of how to use AI, its strengths and weaknesses, and its dangers. In the end, it’s all about access - to knowledge and entertainment. The Internet made it visible; AI makes it personal, intimate, and immediate
This is why AI Pals Engine was made - as a hub and sandbox - to supercharge your desktop, make it legendary, and unlock new possibilities, all in a secure way!
🧩The Main Features
AI Pals Engine is far more than just an AI chat interface. Even without using Text AI, you can craft a truly personalized space with customizable Companions and Desktop Widgets. Just set up the things you want to use! Here are the three key elements that drive it:
🧠AI – Your Powerful, Private, Offline Repository of Knowledge and Entertainment
No cloud. No tracking. Not another subscription. Just you, your thoughts, and your data, which stays on your PC (Optionally: use your own endpoints).
Examples: at its core are the multiple supported Language Models. AI Pals Engine packages additional ones to function, e.g. Text-To-Speech, an Emotion Classifier, a model to compare semantic meaning (used in AI Memory), and more.
AI Based Avatar Expression Behavior:
🤝Pals – More than just Chatbots
These are configurable active companions built from the fusion of Language Model AI and classic Game AI. The Behavior Node System also allows you to configure behavior that goes beyond modifying the chat. It's about automation, interactivity, additional functionality, and making things look awesome.
Examples:
Roleplay AI Pal – A typical chatbot-like Pal that plays a role
Agentic AI Pal – A Pal capable of actually doing things on your PC
Simple Decoration Pal – Just a Pal without AI, like an animated cat idling on your Desktop
Decoration Pal with Behavior – A Pal that acts based on the configured Behavior (e.g. changes its appearance based on the time of day or updates the Note Panel with an RSS feed every minute)
🚀Engine – Your Desktop becomes a living space
The main app serves as a central hub for all your assets (images, models, etc.), to configure your widgets and the active space, to orchestrate all the different technologies used, and of course, to manage your characters, chat history, and more…
Your configured space supports multiple widgets, with multiple Pals, each using different Language Models and Text-To-Speech, each with just the elements you need.
You don’t know how offline AI or AI works? No worries, you can easily use it without being an expert. On your journey or as an expert you will learn to get even more out of AI Pals Engine. With Behaviors and Widgets alone, you can get pretty wild in terms on how you use AI or your PC.
We'll take a closer look at these aspects in the next devlogs!
What Might Not Work for Everyone
It’s important to set expectations. AI Pals Engine is a flexible, experimental sandbox designed for exploration and a variety of use cases. It may not be the best fit for someone who prefers a tool that does just one thing perfectly. Its strength lies in giving you control, freedom, and plenty of functionality to experiment with across different scenarios. In EA the development follows some principles:
Security and Privacy > Bleeding-Edge Tech > UI/UX: Because the underlying technology is evolving quickly, UI polish is not the primary focus yet. Features may shift as development continues, so certain parts of the interface may feel not fully streamlined. However, help text and basic setup wizards are available to guide you where needed.
AI Pals Engine is built for exploration, not oversimplification. The goal is to be user-friendly - but not at the cost of hiding powerful settings or removing commonly used AI terminology. Instead, it encourages understanding what's happening under the hood.
You can absolutely use AI Pals Engine even if you’re new to AI. However, some features - especially advanced behaviors and configuration options - may not feel intuitive right away. It works out of the box but the more you're willing to learn, experiment and explore, the more value you’ll get from it. Tooltips, help pages, and community/AI assistance are always available to support you along the way.
Early access: As stated in the EA section of the store page, AI Pals Engine is a true Early Access. Unlike a fully optimized, marketing-focused launch, this version is meant to focus on your feedback and guide the app's development based on your needs. While the core is polished, many additional features are still being developed, rather than being fully refined from the start.
🔮The Future
I hope to share the official release date soon - it’s not too far away. And since most of the main stuff is basically set, these devlogs are actually meaningful now. The next devlog will follow shortly (likely about Widgets), and let me hear your thoughts. Enjoy your weekend!
Original post

Updates & announcements

What this improves for you:

📝Changelog V0.10.0.0 (publictest Version)

V9.6.0

V9.5.0

V9.4.0 (Current Stable Version)

V<9.4.0

Happy Easter in advance 🐰!

🎙️Voice Activity Mode

How it works

🛠️Workshop & Bundles

Bundles

Looking ahead

🌅What's on the Horizon for AI Pals Engine?

📝Changelog V0.9.1.1 (publictest Version)

🧠AI That Remembers

The Context Window Problem

Memory as Search

Why This Matters

✅ Can I Use Memory in AI Pals Engine?

🔍 How Could We Further Improve the Memory Functionality?

🌅What's on the Horizon for AI Pals Engine?

📝Changelog V0.8.18.0

📝 Changelog V0.8.13.2

🛠️What is Tool Calling?

✅Can I do Tool Calling in AI Pals Engine?

Example 1 – Realtime Info & Grounding

Example 2 – Let your AI do some work by chaining tool calls

Other Examples

🌱Future Plans

🌅What's next for AI Pals Engine?

📝Changelog (summarized from last to V0.8.11.0):

✨ Moving beyond Language Models in AI Pals Engine

🧩 Are they available in AI Pals Engine?

Pal Example

Widget Example

Image Example

🌅What's next for AI Pals Engine?

Changelog V0.8.7.9

🌟 What Are Widgets?

🛠️ Types of Widgets

🎯 The Heart: Panels & Pals

🔍 Deeper Dive

🚀 The Future

❓The Why

🧩The Main Features

🧠AI – Your Powerful, Private, Offline Repository of Knowledge and Entertainment

🤝Pals – More than just Chatbots

🚀Engine – Your Desktop becomes a living space

What Might Not Work for Everyone

🔮The Future

✨ Moving beyond Language Models in AI Pals Engine