Ollama – v0.30.0-rc22

Ollama v0.30.0-rc22 🛠️

If you’re obsessed with running powerful LLMs like Llama 3 or DeepSeek-R1 locally without the cloud headache, Ollama is your best friend. It’s the ultimate toolkit for downloading, managing, and chatting with open-source models right on your own hardware.

This latest release candidate update is all about polishing the experience as we approach the official v0.30.0 stable launch!

What’s new:

• Refinement Step: This version (v0.30.0-rc22) serves as a critical bridge in the pre-release cycle, focusing on fine-tuning the runner performance.
• Stability Focus: As an RC update, the primary goal here is squashing bugs and ensuring everything is rock-solid before the big stable rollout hits your machine.

Keep those engines idling—these frequent RC bumps mean the heavy-hitting new features are just around the corner! 🚀

🔗 View Release

Tater – Tater v81

🥔 Tater v81 — “The Potato Learned Terminal Commands” 💻

Hold onto your hats, tinkerers! Tater just leveled up from a local chat app to a full-blown local AI workspace and runtime system. The potato officially has shell access now, so things are about to get much more interesting (and potentially chaotic)! ⚠️

🚀 Spudex: The New Terminal Workbench

The biggest highlight is Spudex, a brand-new terminal-console workspace built directly into Tater. It’s essentially a “Codex” for your local machine where you can:

• Chat with terminal-capable models.
• Run commands, write scripts, and manage files.
• Inspect outputs and retry/fix failed tasks in real-time.

Pro-tip:* Use containers or isolated folders! Spudex is powerful, so keep those guardrails active to prevent the AI from “reorganizing” your filesystem into abstract art. 🎨

🧠 Hydra Terminal Tools & Smarter Research

Hydra is getting much more capable with new tools like `spudex_run`, `spudex_task`, and `spudex_status`.

• Multi-step execution: Hydra can now handle real terminal tasks instead of just pretending it did them. It even auto-continues unfinished tasks!
• No more “AI Confidence Theater”: Web research is much smarter. Instead of grabbing random snippets, Hydra can now search, inspect pages, and evaluate if it actually found the answer before moving on.

🎙️ Self-Managing Wake Words

Thanks to Firmware v3.0.3, Tater can now change its own satellite wake words!

You can literally say, “Tater, change your wake word to hey computer,”* and it works.

• The new `microwakewordVerba` handles natural spoken phrases better (e.g., “hey tatty” vs “hey-tatty”) and can reset to defaults if things get weird.

📅 Enhanced Personal Core Dashboard

Your dashboard is getting a much-needed productivity boost:

• 7-day outlook context: Get a better grasp of your upcoming schedule.
• Multi-user support: Easily switch between different person profiles.
• Deep Context: Added calendar, action, and subscription context directly into your briefs.

This release marks a massive shift toward Tater being a persistent workflow engine. Happy scripting (and please, use common sense with that shell access)! 🥔⚙️

🔗 View Release

Deep-Live-Cam – 2.7-RC1

🚀 Deep-Live-Cam 2.7-RC1 Update Alert! 🚀

If you’ve been looking for ways to level up your real-time face swapping and live streaming game, this release is a massive leap forward in speed and hardware integration. Whether you’re tinkering with local GPUs or cloud-based diffusion, there is some serious magic happening here.

New Live Modes & Engines:

• Decart Live: You can now stream your webcam directly through Decart’s cloud GPUs using real-time diffusion. Just plug in your API key and choose between the Lucy 2.0 or 2.1 variants! ☁️

FLUX Live (prompt-edit): A powerhouse feature for high-end rigs! This enables real-time, prompt-driven face editing at a smooth 30fps using FluxRT. Warning: You’ll need an RTX 5090 or RTX 6000-class GPU to handle this beast.*

• ReSwapper: A brand new face-swap engine has joined the lineup alongside Inswapper and Hyperswap, giving you more options for your swaps.

UI/UX & Hardware Perks:

• RTX Upscaler: If you have compatible hardware, a new toggle in settings lets you use RTX power to upscale your output beyond the source resolution. 💎
• View Toggles: You can now switch between “Classic” and “Modern” UI layouts with a single click from the status bar.
• Cleaner Live View: The floating player overlay now hides automatically during live streams, giving you a much cleaner viewport for focusing on your content.

Global Accessibility & Polish:

• Full Localization: The interface is officially multilingual! You can switch between English, Chinese, Japanese, French, and Arabic in the settings (just remember to restart the app to apply changes). 🌍
• Improved Window Projection: Resizing during live sessions is now automatic, and a new dedicated close button has been added to the overlay for easier management.

All your favorite tools like RIFE Interpolation, Lip Mask, and Face Enhancer are still baked in and ready to go! 🛠️

🔗 View Release

Text Generation Webui – v4.9

🚀 Major Update Alert: text-generation-webui v4.9 is here!

If you’ve been looking for the “AUTOMATIC1111” experience for your local LLMs, this update is a massive win for efficiency and workflow smoothness. We’re seeing some serious upgrades to how the UI handles web data and inference speed! 🛠️

Smart Web Search Enhancements

• Snippet Support: The `web_search` tool now pulls text excerpts directly from search results. Your model can grab answers without the heavy lifting of parsing entire pages!
• Token Efficiency: The `fetch_webpage` tool is much leaner now, stripping out raw URLs to keep your context window clean and focused on the actual content.
• Polished UI: Enjoy a new loading spinner during searches and much prettier result rendering in your chat interface.

Performance & Inference Tweaks

• MTP Speculative Decoding: New support for `draft-mtpas` is live! It auto-enables when loading MTP GGUFs (like Qwen 3.6 MoE), which can significantly boost generation speeds.
• Live Stats: Monitor your hardware in real-time with live tokens/s tracking and context size monitoring during generation.
• Auto-mmproj Detection: No more manual hunting! The app now automatically detects and selects sibling `mmproj` files when you load a vision model.

UI & Workflow Improvements

• Drag-and-Drop: You can now drag files directly into the chat input for lightning-fast uploads. 📂
• Refined Sidebar: A reorganized sidebar (Mode/Character/Chat) and hidden reasoning controls in simple mode mean much less clutter while you work.
• Electron Upgrades: New “Check for updates” button, a dedicated model directory folder picker, and a handy right-click context menu for easy text copying.

Security & Stability Fixes

• Hardened Security: CORS is now restricted to `localhost` by default, and character name loading has been sanitized to prevent path traversal attacks. 🛡️
• Windows Reliability: Fixed the bug where `llama-server` would hang after the parent process closed on Windows.
• Dependency Refresh: The engine has been bumped up with the latest updates from `llama.cpp`, `ik_llama.cpp`, and `ExLlamaV3`.

🔗 View Release

ComfyUI – v0.22.0

ComfyUI v0.22.0 is officially live! 🚀

If you’re looking for ultimate control over your Stable Diffusion pipelines, ComfyUI remains the gold standard for node-based generative AI. It lets you stitch together complex workflows—from upscaling to inpainting—without writing a single line of code.

This update marks a major milestone within the expanding Comfy-Org ecosystem! While the release focuses heavily on structural improvements, here is what’s driving this new era for your workflows:

• Enhanced Workflow Stability: Expect much smoother interactions between nodes and smarter memory management, which is a lifesaver when you’re running massive, multi-stage generations. 🛠️
• Ecosystem Scaling: As part of the official Comfy-Org push, there is a heavy focus on better integration with custom node managers and more streamlined update processes.
• Performance Tweaks: Ongoing optimizations are rolling out to ensure faster sampling and more efficient execution across your hardware.

Whether you’re running SDXL, ControlNet, or experimenting with LCM models, this version is all about making those complex graphs more reliable and scalable. 🎨

🔗 View Release

Wyoming Openai – Python 3.13 support (0.4.4)

Wyoming OpenAI (v0.4.4) 🚀

If you’re building local voice assistants or tinkering with Home Assistant, this update is a massive win for efficiency and stability! This proxy middleware bridges the Wyoming protocol with OpenAI-compatible ASR and TTS services—allowing you to use heavy hitters like GPT-4o, Whisper, and Kokoro without direct API integration.

What’s New:

• Python 3.13 Support: The project is officially ready for the latest and greatest! Full support is now included across CI, Docker, and packaging metadata.
• Smart STT/TTS Modes: No more wasted resources! If you aren’t using a specific mode (STT or TTS), the client now skips the unused side entirely instead of attempting backend autodetection. This makes startup much cleaner when you only need one direction of audio processing.

Key Fixes & Improvements:

• Fixed Stereo Streaming Timestamps: No more “fast-forward” audio! A bug that caused timestamps to advance too quickly during stereo TTS streaming has been squashed by calculating timing based on PCM frame count. 🛠️
• Robust WAV Header Handling: Fixed an issue where fragmented WAV headers in streaming TTS could leak bytes or cause incorrect audio starts. Now, chunks are buffered until a complete header is ready to be parsed.
• OpenAI Compatibility Patch: A lifesaver for local devs! Recent changes in the `openai` library (v2.34+) were breaking keyless local backends like Speaches and LocalAI. The client now bypasses credential enforcement to ensure your local, keyless setups keep running smoothly without needing fake bearer tokens.

🔗 View Release

Ollama – v0.30.0-rc21

Ollama v0.30.0-rc21 🦬

If you’re running LLMs locally, you know Ollama is the go-to for getting models like Llama 3 and DeepSeek-R1 up and running with zero friction. This latest release candidate is a targeted update focused on stability for our friends over on Windows! 🪟

What’s new:

• Windows Bug Fix: The primary focus of this release is an improvement to exit error logs on Windows. If you’ve been seeing messy or unhelpful logs when closing out sessions, this should make your debugging process much smoother.

It’s a small but mighty update to keep those local environments running clean! 🛠️

🔗 View Release

Lemonade – v10.5.1

🍋 Lemonade SDK v10.5.1 is officially here!

If you’re obsessed with running high-performance LLMs locally without relying on the cloud, this update is a must-have for your toolkit. Lemonade is all about squeezing every bit of power out of your hardware—specifically leveraging NPUs and GPUs (via Vulkan) to make local inference snappy and responsive.

This latest release focuses on keeping your backend integrations rock-solid as the underlying engines evolve:

• llama.cpp Upgrade: The SDK now supports `llama.cpp` build b9213. If you rely on high-performance C++ inference for your GGUF or ONNX models, this is a huge win for stability and speed! 🚀
• AMD/ROCm Optimizations: For my fellow AMD enthusiasts pushing the limits of Ryzen AI or Radeon hardware, we’ve got critical compatibility updates:
• `rocm-stable` has been bumped to build b9211.
• `rocm-nightly` has been updated to build b127.

Time to pull the latest version and keep those local models running smooth! 🛠️

🔗 View Release

Ollama – v0.30.0-rc20

Ollama just dropped v0.30.0-rc20, and it looks like a quick but important fix for the AMD crowd! 🛠️

If you aren’t using Ollama yet, it is the ultimate framework for running powerful LLMs like Llama 3, DeepSeek-R1, and Mistral locally on your own machine. It handles all the heavy lifting of downloading and configuring models so you can focus on building.

What’s new in this release:

• ROCm Build Fix: This update specifically addresses a “cache miss” issue encountered during the ROCm build process.
• Better Stability for AMD Users: If you are running LLMs on AMD hardware, this patch helps ensure your build process is much more reliable and efficient.

If you’ve been tinkering with local models on an AMD GPU, definitely grab this release candidate to keep your environment running smoothly! 🚀

🔗 View Release

Ollama – v0.30.0-rc19

Ollama v0.30.0-rc19 is here! 🛠️

If you’ve been looking for a way to run heavy-hitting models like Llama 3, DeepSeek-R1, or Phi-4 directly on your own hardware without relying on the cloud, Ollama remains the gold standard for local LLM orchestration. It handles all the heavy lifting of model downloading and configuration so you can get straight to prompting.

This latest release candidate (rc19) is part of the ongoing push to refine the local inference experience. While we are still waiting on the full granular changelog for this specific RC, these incremental updates typically focus on:

• Performance Tweaks: Optimization for smoother inference speeds across macOS, Windows, and Linux.
• Stability Fixes: Polishing the REST API and CLI to ensure seamless integration with your custom dev workflows.
• Model Compatibility: Refinements to GGUF support and memory management when swapping between large models.

Keep an eye on this space—as soon as the full breakdown of these specific bug fixes hits the repo, I’ll let you know exactly what’s under the hood! 🥔✨

🔗 View Release