GoKawiil
Tech News
Ollama, the open-source platform for running large language models locally on macOS, Linux and Windows, has released an update adding support for a new 65B 'Ripple Preview' model aimed at agentic coding tasks. The release also cuts memory usage for Qwen3.8 Flash Next by about 27 GiB on Apple Silicon, improves Claude Code integration, and adds CUDA 13.5 support for Windows on ARM.
Opera announced it will ship open-weight AI models directly inside its browser, powered by the newly proposed WebAI web standard backed by Google and Microsoft. Users will be able to run AI queries locally without installing separate tools like Ollama, and the feature is opt-in, with nothing downloaded unless the user chooses to.
Ollama has added text-to-speech support to its Mac app, starting with Phonon, a family of small speech models from Gradium. The feature generates audio in real time as text is produced, and Ollama says Windows and Linux support is coming soon. Ollama reports its MLX-based implementation runs 35% faster than Gradium's own version in tests on an M5 Max MacBook Pro.
A blogger has published a tutorial explaining how to set up systemd-nspawn containers, describing the tool as a lightweight, namespace-based alternative to Docker that uses a writable root filesystem rather than prebuilt images. The post follows up on an earlier mention of isolating self-hosted Ollama and Open WebUI instances using the technology, and was prompted by a footnote that grew into a separate explainer on why Docker wasn't used instead.
A new desktop application called Linkit was released as a free, open-source bookmark manager featuring AI-generated summaries, semantic search, and a knowledge graph. It stores data locally by default in SQLite with optional cloud sync, supports one-click import from Chrome, Firefox and Raindrop, and works with any OpenAI-compatible API or local models like Ollama. The app is available on macOS, Windows and Linux, with macOS installable via Homebrew.
An independent developer released qwen3-4b-deslop, a fine-tuned version of Qwen3-4B-Instruct trained on a MacBook Air using 161 example answers to remove stylistic habits like em dashes, 'not X, but Y' constructions, and lists of three. The model is available as a 2.5GB download runnable via Ollama, and on 40 held-out test prompts its measured 'slop score' fell from 121 to 12.5 per 1000 words while maintaining accuracy on 30 factual questions.
Apple researchers detailed a system called Minions in an arXiv paper, which harnesses idle Apple Silicon Macs across the company's network to process AI inference workloads. The system splits jobs—embedding tasks and LLM text generation via Ollama—into subtasks that idle Macs pull and execute, cutting wait times from over 10 hours in GPU queues to minutes.
Reddit user markpronkin bought a rusty, untested nine-GPU crypto mining rig found in a damp garage for about $35 (3,000 rubles) after the seller refused a pre-purchase test. The rig's nine Nvidia P106-100 cards, each with 6GB of GDDR5, provided 54GB of combined VRAM and worked on the first attempt, running Alibaba's 35-billion-parameter Qwen3.6-35B-A3B model in Ollama at over 30 tokens per second.
Scientists including Florida Atlantic University neuroscientist Ian Gaudet are increasingly running large language models locally on their own computers rather than using cloud-based chatbots such as Claude, Gemini or ChatGPT. Tools like Ollama, LM Studio and DEVONthink let researchers download AI models to search and organize their literature and data offline, avoiding the risk that cloud providers could use their inputs for training. Users cite concerns over data protection, reproducibility and cost as reasons for going local, though rising hardware costs and the need to verify AI output remain obstacles.
Lumen's Black Lotus Labs reports that a campaign called Canto Incognito has infected more than 3,400 servers since April 2026 using malware dubbed PoeLLM, which deploys XMRig and Iron cryptominers linked to Kryptex mining infrastructure. The malware hides its command-and-control addresses inside words of a two-stanza poem posted on GitHub, edited 11 times to redirect infected machines to new servers, and primarily targets exposed AI/LLM tools like LiteLLM, Ollama, Gotenberg, Gitea, and possibly Ivanti Sentry.
Black Lotus Labs reports that a malware strain called PoeLLM has compromised more than 3,400 exposed AI servers, with up to 800 infected systems active on a single day. The malware, an ELF file named libgcrypt, derives its command-and-control IP address by extracting keywords from a poem hosted in a GitHub repository, updating the poem 11 times since April to rotate C2 addresses.
Parrot is a free, open-source meeting recorder for Mac that captures audio your computer already plays, rather than joining calls as a bot. It works with Zoom, Meet, Teams or any audio source, offers searchable transcripts via a local command palette, and can integrate with tools like Claude Desktop, Cursor and Codex. Users can run it fully offline using Whisper and Ollama, or opt into cloud AI providers like Groq or Deepgram for transcription, paying those providers directly.