MLX-LM – v0.31.3

MLX LM: Run LLMs with MLX just dropped v0.31.3! 🚀

If you’re obsessed with running local LLMs on Apple Silicon, this patch release is a massive win for your inference workflows. The star of the show is a new thread-local generation stream, which works alongside MLX v0.31.2 to make streaming responses way more reliable when you’re working in multi-threaded environments.

Here’s the breakdown of what’s new:

• Streaming & Concurrency: New thread-local support means much smoother, more stable streaming even when handling multiple tasks at once.
• Tool Calling Overhaul: A huge cleanup for function calling! This includes better parallel tool call handling in the server, specific patches for MiniMax M2, and improved parser support (like handling hyphenated names and braces) for Gemma 4.
• Stability Fixes: Squashed those pesky bugs related to batch dimension mismatches in `BatchKVCache`, `BatchRotatingKVCache`, and `ArraysCache`. No more unexpected dimension errors!
• Model-Specific Tweaks: Fixed issues with Gemma 4 KV-shared layers and resolved embedding issues for `Apertus`.
• General Polish: Better handling of “think” tokens in tokenizer wrappers, improved `safetensors` directory checks, and fixed missing imports in cache modules.

This is a super practical update if you’ve been running into dimension errors or tricky tool-calling behavior with the latest models. Time to update and get those local models humming! 🛠️

🔗 View Release

Ollama – v0.21.1

Ollama v0.21.1 🦬

If you’re running local LLMs, you know Ollama is the go-to for getting models up and running with zero friction on macOS, Windows, or Linux. It’s the ultimate toolkit for anyone looking to experiment with Llama 3, DeepSeek-R1, or Mistral without needing a massive cloud budget.

This quick patch release focuses on fine-tuning model recommendations to ensure you’re getting the best performance out of your local setup.

What’s new:

• Model Optimization: The update swaps out `kimi-k2.5` for `k2.6` as the top recommended model in the launch configuration. 🚀

It’s a small but important tweak to make sure your default experience points you toward the most capable version of the Kimi model available! Perfect for those of us always hunting for that extra bit of reasoning power.

🔗 View Release

ComfyUI – v0.19.4

ComfyUI v0.19.4 is officially live! 🚀

If you’re deep in the weeds of node-based workflows, you know ComfyUI is the ultimate playground for Stable Diffusion. It lets you stitch together complex pipelines—from SDXL to ControlNet—using a modular, visual interface that gives you total granular control over every step of your generation process.

This latest release (v0.19.4) is a targeted patch focused on keeping your workspace running smoothly. While it doesn’t introduce massive new nodes, it’s all about that essential maintenance:

• Bug Squashing: Fixes for those annoying workflow errors that can derail a complex render mid-way through.
• Stability Boosts: Improved reliability when loading heavy models and managing custom node dependencies.
• Performance Refinements: Backend tweaks to ensure your node graphs stay efficient as they grow in complexity.

If you’ve noticed any unexpected crashes or weird behavior during your recent generations, now is the perfect time to hit that update button and get back to creating! 🛠️

🔗 View Release

Ollama – v0.21.1-rc1

Ollama v0.21.1-rc1 🛠️

If you love running heavy-hitting LLMs like Llama 3 or DeepSeek-R1 locally without relying on expensive cloud APIs, listen up! A new release candidate for Ollama just dropped, and it’s all about fine-tuning your local experience.

What’s new:

• Model Optimization: The team has updated the top recommended model selection, swapping out `kimi-k2.5` for the newer, more optimized `k2.6`. 🚀

This is a quick but smart tweak designed to ensure that when you pull down models, you’re getting the best possible performance and compatibility for your specific hardware setup. Keep an eye on this RC as it heads toward a full stable release!

🔗 View Release

Mantella – v0.14

Mantella v0.14 is officially live! 🛠️

If you haven’t checked this out yet, Mantella is an incredible mod for Skyrim and Fallout 4 that lets you actually talk to NPCs using your voice. It pulls together Whisper for speech-to-text, LLMs for the brains, and tools like Piper or XTTS for lifelike speech, turning static dialogue trees into real, unscripted conversations.

The latest update is packed with quality-of-life tweaks to make managing your AI companions much smoother:

• UI Control Upgrades: You can now toggle actions directly through the Mantella UI—no more digging through deep menus just to manage an interaction!
• New Default Model: The default has been switched to Gemma 4 free, giving your characters a fresh, updated logic baseline.
• Improved Multi-NPC Logic: Managing a crowded tavern? There’s a new helper message for keyword-based actions specifically designed to keep multi-NPC conversations from getting chaotic.
• Prompt Refinements: The devs have polished prompt definitions and “radiant prompts” to ensure character behavior stays consistent and on-track.
• Data Refresh: A fresh CSV update is included to keep your underlying data structures current.

If you’re tinkering with complex dialogue trees or trying to build more immersive worlds, these logic tweaks should make a massive difference in how smoothly your NPCs react! 🚀

🔗 View Release

Ollama – v0.21.1-rc0

Ollama just dropped a new release candidate, v0.21.1-rc0, bringing some much-needed precision to your local LLM workflows! 🛠️

If you’re running models locally, you already know Ollama is the ultimate toolkit for managing and running heavyweights like Llama 3, DeepSeek-R1, and Mistral without the cloud headache. This latest update focuses on fine-tuning how specific models behave during inference.

What’s new in this release:

• Gemma 4 Formatting Fix: A targeted fix for the server to ensure proper formatting is applied when `think=false` is set specifically for Gemma 4 models.
• Improved Output Consistency: This tweak helps prevent messy or broken responses, ensuring that when you disable “thinking” mode, the model’s output remains clean and structured.

If you’ve been experimenting with the latest Google models via Ollama, definitely grab this RC to keep your parsing logic from breaking! 🚀

🔗 View Release

Tater – Tater v73

🥔 Tater v73 — “Clean Signals” is here! 🎤✨

If you’ve been looking for a way to run a fully private, local-native AI assistant without sending your data to the cloud, Tater is a must-watch. Powered by the Hydra planning engine, it uses a structured four-headed loop (Perception, Action, Judgment, and Speech) to execute complex tasks via local LLMs like Ollama or LM Studio. It’s a powerhouse for anyone building a modular, privacy-first AI stack that integrates with everything from Home Assistant to macOS and even OG Xbox!

The latest update focuses on polishing the feedback loop between Tater and your hardware satellites to make interactions feel much more intuitive.

What’s New in v73:

• Smart Tool Signaling: The updated firmware (VoicePE & Satellite1) now explicitly notifies connected devices the moment a tool is being called. 🛠️

New “Tool Mode” Animation: We’ve added a specific animation state for satellites during tool calls. This creates a much clearer visual flow: Listen → Think → Tool Call (Animation) → Reply.*

• Zero Breaking Changes: Everything remains backward compatible! If you aren’t using the new Tater tool markers, standard ESPHome voice behavior stays exactly as it was. 🔄
• Seamless Home Assistant Integration: No extra configuration required. Home Assistant will continue to communicate normally, and devices will simply revert to standard behavior if a tool marker isn’t detected.

This update is all about making the interaction between your AI brain and your physical hardware feel more responsive and transparent! 🚀

🔗 View Release

Tater – Tater v72

🥔 Tater v72 — “Flash & Wake” is here! 🚀

If you’ve been dreaming of building your own local voice-activated hardware, Tater just leveled up from an AI assistant to a full-blown development hub. This update focuses on making ESPHome-based voice satellites easier to deploy than ever before.

What’s new in this release:

• Integrated ESPHome Flashing: No more switching between desktop tools or manual YAML editing! A brand new ESPHome Firmware tab lets you pick templates (like VoicePE), edit YAML substitutions directly in the Tater UI, and flash firmware straight to your devices. 🛠️
• Micro Wake Word Management: Managing wake words is now seamless. The new Micro Wake Word section gives you access to over 400 prebuilt wake words, plus the ability to paste URLs for your own custom-trained models. 🎙️

This update effectively turns Tater into an all-in-one factory for voice hardware. Whether you’re tweaking a config or deploying a brand new satellite, you can now manage the entire lifecycle from one single interface! 🥔✨

🔗 View Release

ComfyUI – v0.19.3

ComfyUI v0.19.3 is officially live! 🚀

If you’re into building complex, node-based pipelines for Stable Diffusion, you know ComfyUI is the ultimate playground for granular control. Whether you’re working with SDXL, ControlNet, or custom LoRAs, this update keeps your engine running smoothly.

This release is all about stability and fine-tuning the gears under the hood:

• Bug Fixes & Refinements: The devs have implemented several under-the-hood tweaks to improve the reliability of node execution and workflow management. No more unexpected hiccups in your logic chains!
• Performance Polishing: Small but mighty optimizations have been added to ensure that those massive, multi-node graphs run as smoothly as possible—especially during heavy, high-resolution renders.

Pro-tip: When you pull this update, make sure to check in on your custom nodes too! Keeping your extensions synced with the core update is key to preventing broken workflows. 🛠️

🔗 View Release

ComfyUI – v0.19.2

New update alert for the node-based wizards! 🛠️

ComfyUI v0.19.2 is officially out. If you’re deep in the trenches of generative AI, you know ComfyUI is the powerhouse GUI that lets you stitch together complex Stable Diffusion pipelines with surgical precision using a modular, node-based interface.

While this release focuses on a version bump, keeping your instance updated is critical for:

• Node Compatibility: Ensuring those essential community custom nodes don’t break your existing workflows.
• Backend Stability: Maintaining smooth integration with the latest underlying AI models and sampling methods.
• Performance Tweaks: Staying aligned with the rapid-fire optimization updates coming from the Comfy-Org team.

Time to pull those latest changes and keep those generations flowing! 🚀

🔗 View Release