Three of this week's four tools run on hardware you already own

The Lens

Unsloth trains open models roughly twice as fast on about 70% less VRAM, which is the difference between renting a datacenter GPU and fine-tuning on the card already in your desktop. Fine-tuning means taking an existing open model and teaching it your specific task or data. Unsloth is now two things: Unsloth Core, the Python library and CLI, and Unsloth Studio, a local web UI you launch with a single command for downloading, running, and comparing models. It runs on NVIDIA RTX 30/40/50 and Blackwell, on AMD across Windows, WSL, and Linux, and on macOS for both training and inference. CPU handles chat and data prep. Multi-GPU is supported. The reinforcement learning path claims 80% less VRAM for GRPO, and long-context training reaches 500K tokens. Studio also handles inference: GGUF, LoRA adapters, safetensors, tool calling, and API endpoints. Free, with no paid tier and no commercial restriction on Core. Solo: this is the default way to fine-tune on consumer hardware. Teams: same answer, and multi-GPU scales it up. The catch is the license split. Core is Apache 2.0 and carries no obligations. Studio's UI components are AGPL-3.0, and AGPL follows you if you embed that UI into a product you ship or host for others. For personal training runs it is irrelevant. For a company building a product on top of it, know which half you are standing on before you get far.

The Lens

ToolJet gives you a drag-and-drop builder for internal tools (admin panels, dashboards, CRUD apps) that connects to your databases, APIs, and SaaS tools. It is the open source, self-hostable take on Retool. Connect Postgres, MySQL, MongoDB, REST, GraphQL, Google Sheets, and dozens more, drag components onto a canvas, wire them to queries, and you have a working tool. AGPL-3.0. Self-hosting runs on Docker or Kubernetes, and ToolJet says a first deploy usually takes under 30 minutes. Read the fine print on what free means. The pricing page's free self-hosted plan lists 2 builders, 50 end users, and 2 apps, and the README quickstart pulls the enterprise image. A Community Edition image exists under AGPL, but we could not find ToolJet stating in writing whether those caps apply to it. Cloud pricing, billed annually: Basic $23 per builder, Pro $79, Team $199. Month to month, Pro is $99 and Team is $249. Pro caps at 50 end users and Team at 100. Past that, the only plan is Enterprise, from $3,000/mo. OIDC, LDAP, and SAML single sign-on, audit logs, and white-labeling start at Team. Solo and small: free tier, or Pro past 2 apps. Medium: Team. Large: price out appsmithorg/appsmith and Budibase/budibase first. The catch: AGPL means modified versions served to outside users must be open sourced. Team audit logs only keep 14 days. And two builders who need SAML are paying $400 to $500 a month.

The Lens

H3.c generates video with synchronized audio locally on a Mac. It's a from-scratch C inference engine for MiniMax's H3 model, written by the creator of Redis, with an interactive REPL that previews output right in your terminal. The code is MIT and a pleasure to read; local, private video generation with no cloud API is the pitch. The hardware is the gate. You want Apple Silicon with serious unified memory: peak usage runs into the tens of gigabytes depending on canvas size, plus about 37 GB of disk for the checkpoint. An SSD-streaming mode drops memory use to around 2 GB, but at a steep speed penalty at small canvas sizes. Build is a simple make with FFmpeg on PATH. This is for Mac owners with serious unified memory who want local generation, and for anyone who wants to read a modern diffusion pipeline in C. ComfyUI is the mainstream local workbench; Runway and Pika are the hosted alternatives. The catch: the MIT badge covers only the code. The model weights ship under MiniMax's custom community license, non-OSI, with an application process for US, EU, UK, and South Korea users, and the engine is useless without them. The author also frames the repo as working vertical slices, not a finished product, and canvas sizes and frame counts have sharp constraints.

The Lens

oMLX runs large language models on your Mac and manages the whole thing from the menu bar. Pick a model, hit start, and you have a local OpenAI-compatible API at localhost:8000 that Claude Code, Codex, or any OpenAI client can point at. Apache 2.0, no account, no paid tier. The engineering underneath is more serious than the menu bar suggests. Continuous batching handles concurrent requests, and a two-tier KV cache keeps hot blocks in RAM and spills cold ones to SSD, so a long context survives a server restart instead of being recomputed from scratch. One process serves text models, vision models, OCR, embeddings, and rerankers, with model pinning, per-model idle timeouts, sampling settings, and a memory ceiling all editable from the admin dashboard. Solo developers running a local coding assistant get the most out of it, and a small team can share one big Mac over the API. Ollama (ollama/ollama) is easier to hand to a teammate and runs on any hardware. llama.cpp (ggml-org/llama.cpp) gives you more low-level control. oMLX wins specifically on Apple Silicon. The catch: Apple Silicon and macOS 15 or newer, full stop. And the fast native Metal kernels for the GLM and MiniMax families do not build from a plain pip install. Without full Xcode you silently fall back to a path the project measures at roughly 30x slower on prefill. Use the official DMG, which ships them precompiled.