Three tools orbit one closed AI model, plus a free Photoshop for Mac

The Lens

Laya answers typed questions about text without generating any: pick one of these labels, give this a score, yes or no. Hand it a support ticket or an email and it returns probabilities in a single pass. It's an open-weight rival to TypeSafe's hosted Jev model, and it copies Jev's API, so an existing client switches by changing one URL. Apache 2.0, code and weights. It runs on CPU. The README puts one question at roughly 200 to 460 milliseconds on CPU and about 33 on a T4 GPU. There are three checkpoints of 322M to 421M parameters, one covering 100+ languages, plus a built-in HTTP server and MCP server. Jev charges $42 per billion input tokens; this costs whatever hardware you already have. mizorewww/laya-mlx ports it to Apple Silicon. Teams routing tickets, triaging email or gating agent actions at high volume: self-host it and fine-tune it. The README is blunt that out of the box it's near chance on real workflows, and says to treat it as a fast base to specialise, not a zero-shot engine. A fine-tune takes about 4-5 hours on Kaggle's free GPUs. The catch: its Jev comparisons use Jev's published numbers, not side-by-side runs, and with many options Jev wins clearly. Confidence scores behave differently too, so thresholds tuned on Jev need re-tuning.

The Lens

fast-jev-compaction changes what happens when a long Claude Code session runs out of room. The built-in behavior summarizes old conversation, which loses detail. This plugin asks Jev, a decision model from TypeSafe AI, which old tool calls and outputs are stale, deletes only those, and keeps every user and assistant message word for word. MIT licensed. Install goes through the Claude Code plugin marketplace, straight from the repo. It needs an environment flag for function hooks, which the README calls an early-access Claude Code feature, plus a TypeSafe API key. Jev bills $0.042 per million input tokens and is waitlisted. When Jev errors out or does not cut enough, the plugin falls back to the normal summary. Solo developers living in long Claude Code sessions are the audience, once they have Jev access. Teams should think harder, because the plugin sends your session transcript to TypeSafe for scoring. The catch: do not run the npm install line in the README. As of September 21, 2026, the fast-jev-compaction name on npm belongs to a different account publishing its own fork with extra command line tools. Nothing in it looked malicious on a static read, but it is not the author's package. The marketplace path pulls from the repo and is unaffected. Token sizes are also estimated from character counts, and users report the hooks not loading on some Claude Code versions.

The Lens

laya-mlx runs Laya decision models natively on Apple Silicon. Laya is an open-weights model from Convai Innovations that answers multiple choice, rating, and yes or no questions with probabilities instead of generated text. It is the open counterpart to TypeSafe AI's closed Jev API. This port drops PyTorch and runs on Apple's MLX, returning a short decision in roughly 7 to 14 milliseconds on an M3 Max by the author's own measurement. Apache 2.0. Setup is pip install laya-mlx on Python 3.11 or later and macOS 14 or later. Weights download from Hugging Face on first load, and peak memory for a short question stays under 1 GB. No cloud API and no account. It covers inference and conversion only. Training and fine-tuning stay in the upstream Laya project. Free at every size. Use it for local classification, routing, or guardrail checks on a Mac where a full LLM call is too slow. TheoLeeCJ/SemIf gets a similar interface out of general open models and also has an MLX backend. The catch: this is an unofficial port from a young pseudonymous account, at version 0.1.0 with five commits. Context tops out at 512 to 1,024 tokens. And an open issue reports that swapping the order of the options flipped the answer 27 to 33 percent of the time on a Chinese benchmark. Test option-order sensitivity before trusting it.

The Lens

Compositor is a free, native Mac image editor built to feel like Photoshop, down to the keyboard shortcuts, so the muscle memory carries over. Layers, folders, masks, clipping masks, adjustment layers for curves and levels, a clone stamp, spot healing, content-aware fill, subject selection, and background removal all work today. MIT licensed, no account, no subscription. Install is a signed and notarized DMG from GitHub Releases with automatic updates. The hard requirement is macOS 26.5. Anything older will not run it, and an open issue asks for macOS 15 support. It is Mac only. It imports JPEG, PNG, HEIC, TIFF, and as of version 1.2 Photoshop PSD files, keeping folders, masks, and some blend modes editable. Solo creators and small teams who composite and retouch photos: try it before renewing Adobe. Anyone who trades PSDs with clients should wait. On Windows or Linux, GIMP and Krita are the established open source options, and penpot/penpot covers UI design, not photos. The catch: it was five days old when we looked, shipping several releases a day. PSD support is import only, 8-bit RGB only, no CMYK, and PSD text layers turn into pixels. There is no PSD export at all. The 1.2 release notes admit layer effects were being lost on save until that fix. Keep backups of anything that matters.