Hob (by Frigade)
New and early (v0.2, first release Oct 26, 2026): free desktop app for Apple silicon Macs, Windows and Linux that works on your local files like the ChatGPT or Claude desktop apps, but runs open-weight models (Gemma, gpt-oss, Qwen) on your own computer and picks one that fits your hardware. Caveat: The Terminal (Oct 27, 2026) notes the GitHub README lists no hardware requirements and describes no permission model for the files it edits; the hob.chat FAQ gives memory guidance and says it asks before changing files in a folder you share.
Research notes
How we research- Official site loadsChecked Oct 27, 2026: hob.chat
- Pricing checked on the vendor siteSource: https://github.com/FrigadeHQ/hob
- Free / DIY pathFree download; models run on your computer and it works offline
Listed because it fits the Local AI category. Listings are free; sponsored placements, if any, are labeled and never change rankings.
Alternatives
Ollama
Local AIRun open models locally with one command and a local API.
LM Studio
Local AIDesktop app to download and chat with local models (GGUF/MLX).
llama.cpp
Local AIC/C++ LLM inference engine behind many local AI apps.
MLX
Local AIApple's array framework for fast ML on Apple silicon.
Open WebUI
Local AISelf-hosted ChatGPT-style web UI for Ollama and other backends.
vLLM
Local AIHigh-throughput LLM serving engine for NVIDIA GPU servers.
Mac Studio (hardware)
Local AIApple desktop with large unified memory, popular for local LLMs.
NVIDIA DGX Spark (hardware)
Local AIDesktop Grace Blackwell AI system in 64GB or 128GB unified memory configurations; the 64GB version is sold only through OEM partners.
GPT4All (Nomic)
Local AIDesktop app for running local LLMs privately.
AnythingLLM
Local AIAll-in-one local/self-hosted AI app with RAG and agents.
LocalAI
Local AIOpen-source, OpenAI-compatible API server for running models locally.
Msty
Local AIDesktop app for local and cloud LLMs.
Llama (llama.app)
Local AIFree open-source tray app from ggml-org that runs Hugging Face models locally on llama.cpp with an OpenAI-compatible API. On Mac, and on Windows 11 as Llama Windows (announced by Hugging Face Oct 13, 2026), including Windows on Arm.
Mac mini (hardware)
Local AICompact Apple Silicon desktop for small local-model setups.
LibreChat
Local AIOpen-source, self-hosted ChatGPT-style platform: one interface for local models (Ollama, LocalAI) and cloud models, with agents, MCP, document search and SSO for a private team chat.
oMLX
Local AIFree, open-source MLX model server for Apple Silicon Macs, with a menu-bar app, multi-model serving, an SSD-backed KV cache, and OpenAI- and Anthropic-compatible APIs. Advanced: best for Mac Studio owners wiring up coding or agent tools; pre-1.0 releases from a single maintainer.
Paperless-ngx
Local AISelf-hosted document archive: scan, OCR, index and search receipts, permits, lien waivers and COIs.
Docling
Local AIConverts PDFs like specs and submittals into clean text/Markdown that local AI models can read.
whisper.cpp
Local AIOffline speech-to-text for site-walk voice memos and meeting recordings, run on your own machine.
BetterWispr
Local AIFree, open-source Mac dictation: hold Option+Space in any app and talk, and speech is transcribed on your Mac (Apple speech, Parakeet or Whisper models). Built-in meeting notes. New (v0.1.x), Mac only.
Shiyin
Local AIFree, open-source meeting-notes app for Mac. Records in-person meetings and Zoom, Meet or Teams calls without a bot, transcribes and identifies speakers on your Mac, and writes summaries and to-dos that link back to the recording. Point summaries at Ollama or LM Studio to stay fully offline. Needs macOS 26 on Apple Silicon; small solo project.
exo (Exo Labs)
Local AIFree, open-source app that links your own machines (Exo's site names Mac Studio and DGX Spark) into one local cluster so you can run bigger open models than one box can hold, managed from one dashboard. Connects over Thunderbolt or USB-C and serves OpenAI-, Claude- and Ollama-style APIs to the tools you already use. The macOS app needs macOS Tahoe 26.2 or later on Apple silicon; Linux installs from source. Exo also has a contact-sales 'For business' option with no public prices.
Lemonade Server
Local AIFree, open-source local AI runtime with a GUI, CLI and API that serves chat models plus vision, image, speech, transcription and embeddings on your own GPU or NPU. Installers for Windows, macOS, Docker and Linux; a good fit for Windows and AMD PCs.
Recursive Flint
Local AIFree open-weights coding model (40B mixture-of-experts, about 2.5B active) built to run on your own Mac or NVIDIA machine. The Apple Silicon INT4 build is about 22.6 GB; the NVIDIA GGUF card recommends 48 GB of memory. Speeds and benchmarks so far are Recursive's own.
Open Notebook
Local AIFree, open-source, self-hosted alternative to Google NotebookLM. Load PDFs, web links, YouTube videos and documents into a private notebook, then chat with them, summarize them and generate podcast-style audio. Works with 20+ model providers, including local servers like Ollama, LM Studio and vLLM. Runs with Docker Compose.
Off Grid AI
Local AIFree, open-source app that runs AI models on the phone and computer you already own (chat, documents, photos, voice, image generation) with no account and no internet needed, which suits crews without signal. The desktop app also serves a local OpenAI-compatible API. License split: the mobile app (iOS/Android) is MIT; the desktop core is AGPL-3.0. Optional paid Pro tier. Launched on Product Hunt Oct 27, 2026; no independent press yet.
Strata
Local AIVery young, mostly one-developer open-source inference engine (first commit Sept 2026) that runs large mixture-of-experts models such as Qwen3.8-Flash-Next, and DeepSeek V4.1-Flash on high-RAM machines, on consumer PCs by spreading the model across GPU, CPU, RAM and SSD, with OpenAI- and Anthropic-compatible APIs on localhost. Installers for Windows, Linux and Apple silicon Macs; the README says DeepSeek on a Mac needs 64 GB or more and is text only. Caveat: speed figures in the README are the developer's own; LinuxCompatible (Oct 25, 2026) noted skepticism over heavy online promotion and no independent minimum-spec benchmark.