Last week my Mac got down to 3.7 GB of free space. Coding agents had spun up dozens of worktrees, each with its own node_modules, .build, and DerivedData. So I did what everyone does in that moment: I started deleting folders by hand, more or less at random.
One of those folders was a 9.3 GB worktree that held the only copy of a data set I needed. Oops!
So I made Plonge 🧽 It finds the caches that dev tools and coding agents leave behind, shows you what it found, and deletes only what's safe to go.
Yesterday it took my laptop from 58 GB free to 98 GB. The latest release has rules for every core tool that mise installs, plus an MCP server so your agent can clean up after itself.
(Why "plonge"? In a French kitchen, the plonge is the dishwashing station. If @jdxcode's mise-en-place is how you set up, plonge is how you clean up.)
```
brew install mattt/tap/plonge
```
github.com/mattt/plonge
Microduck can now be your pet in @ChatGPT. Build your own and delight in its squawks, stomps, and faceplants as Codex does its thing 🦆
mattt-pollen-codex-pets.stat…
Over the last 30 days, Sosumi.ai has served tens of millions of requests to coding agents looking up Apple docs, from about 6,900 (nice) unique IPs/day.
But does it actually help? I built an eval to find out: 41 Swift tasks, mostly new APIs in the OS 27 SDKs, checked by the compiler and per-task rules.
Without docs, Opus, GPT-6.1 Sol, and Sonnet pass about a third of them. With Sosumi, they pass 98%, 97%, and 94%.
Compared to built-in web search, Claude models score 12 to 30 points higher with Sosumi, at about half the cost or less. GPT does slightly better with web search (99%), but at 3× the cost.
Running the eval also turned up a handful of fixes, now live (one framework page was 3 MB 😬). The eval is open source: github.com/NSHipster/sosumi.…
AnyLanguageModel 1.2 is out! 🎉
This release brings dynamic profiles from Foundation Models 27 to earlier OS versions and to other models, so a session can now change its model, instructions, and tools as it goes.
Huge thanks to @llqoli for this work, and to @jaleel_akbashev, @mergesort, and ﹫jihtsan.
github.com/huggingface/AnyLa…
We’ve been hard at work trying to make models work with harnesses better. The best thing @yb2698 found: “harness adapters”. Small LoRAs trained with RL inside the actual agent harness they’ll be used with.
Same base model. Different adapter for pi, opencode, hermes, etc. Why?
Simply because the same model behaves very differently depending on the tools, prompts and control flow around it. Instead of asking one set of weights to be optimal everywhere, we can specialize a tiny part of the model for each environment.
Early results are pretty encouraging: on held-out tasks, the adapters make fewer tool calls than the base model while keeping or improving accuracy.
And because they’re LoRAs, the operational side is surprisingly cheap too.
For one LFM2.5-24B-A2B experiment, we evaluated 9 checkpoints / 1,350 agent sessions on a single H200 in under an hour. The total: ~$5!
Merging and serving every checkpoint separately would have cost around $90.
There’s another important piece: RL agents will eventually try to hack the reward.
So every tool call goes through `safeagents`, a deterministic rule-based guard that blocks things like editing tests, tampering with rewards or bypassing the sandbox before they execute.
And the whole loop can start from real usage:
your coding sessions -> verifiable environments -> RL -> a better adapter for your agent.
The model doesn’t have to change completely.
Sometimes it just needs to get better at the harness it actually lives in.
huggingface.co/blog/harness-…
Apple's Core AI models repo is now accepting third-party contributions 🥳
If you have developed some cool ports, this is the best place to share your recipes with the community. Looking at you @JackdeS11@mattt@kevinqz 😂
github.com/apple/coreai-mode…
I also opened a feature request for Paseo to add a microphone setting, since this seems like something the app should handle: github.com/getpaseo/paseo/di…
Dictation is half of how I use this. But when I put on AirPods, macOS makes them the default input, and the moment I dictate, whatever's playing gets loud and drops to walkie-talkie quality.
So I built a small plugin that keeps Paseo on the mic you pick: github.com/mattt/paseo-micro…
I made Paseo Micro: a plugin and macOS app that turns a @work_louder Creator Micro 2 keypad into a controller for my Paseo coding agents.
github.com/mattt/paseo-creat…
I've been using @moboudra's Paseo for the past month, and it's been a great fit. I was getting so much done that noticing which agents were waiting on me had become the bottleneck. Then I saw @Thom_Wolf switching between 36 sessions with a tiny Work Louder keypad, and impulse-bought one.
There were probably easier, definitely more responsible ways to solve this. But buying a cool toy seems Pareto optimal (right?). I hadn't even checked whether it was compatible. Just took a leap of faith: "Claude, take the wheel."
Fortunately, @chacon had reverse-engineered the keypad's USB HID protocol and documented it in micro-manager. Following his work, I had a working plugin within an hour. I've spent the past week polishing it, and I'm really happy with it.
The translucent key glows green when an agent is waiting for me. I tap it, and Paseo opens to that conversation, even from the background. The double-wide key is hold-to-talk. So I see green, tap, read, and hold to respond by voice.
The top rows have keys to navigate back and forth. The dial scrolls the conversation (press while scrolling to go faster), and the joystick opens a radial menu of common responses. But the next-agent key and dictation get about 90% of my use.
(The photo is AI-generated from my own pictures of the keypad, because my desk was a mess.)
iMCP 1.8.2 fixes a security issue: in 1.8.1 and earlier, another process on your Mac could intercept an MCP client's session with iMCP. Please update with "Check for Updates..." or brew upgrade --cask imcp, then follow the steps in the release notes:
github.com/mattt/iMCP/releas…
I made Paseo Micro: a plugin and macOS app that turns a @work_louder Creator Micro 2 keypad into a controller for my Paseo coding agents.
github.com/mattt/paseo-creat…
I've been using @moboudra's Paseo for the past month, and it's been a great fit. I was getting so much done that noticing which agents were waiting on me had become the bottleneck. Then I saw @Thom_Wolf switching between 36 sessions with a tiny Work Louder keypad, and impulse-bought one.
There were probably easier, definitely more responsible ways to solve this. But buying a cool toy seems Pareto optimal (right?). I hadn't even checked whether it was compatible. Just took a leap of faith: "Claude, take the wheel."
Fortunately, @chacon had reverse-engineered the keypad's USB HID protocol and documented it in micro-manager. Following his work, I had a working plugin within an hour. I've spent the past week polishing it, and I'm really happy with it.
The translucent key glows green when an agent is waiting for me. I tap it, and Paseo opens to that conversation, even from the background. The double-wide key is hold-to-talk. So I see green, tap, read, and hold to respond by voice.
The top rows have keys to navigate back and forth. The dial scrolls the conversation (press while scrolling to go faster), and the joystick opens a radial menu of common responses. But the next-agent key and dictation get about 90% of my use.
(The photo is AI-generated from my own pictures of the keypad, because my desk was a mess.)
swift-transformers 1.4.0 has just been released 🚀
Focus: tokenization correctness, performance, and alignment to transformers. PRs from 15 different contributors, 10 of them new. Thank you! 🙏
github.com/huggingface/swift…
iMCP 1.8.1 fixes a security issue: in 1.0.1 through 1.8.0, an MCP client could call iMCP's tools without your approval. Please update with "Check for Updates..." in iMCP's menu, or with brew upgrade --cask imcp.
github.com/mattt/iMCP/securi…
Happy Halloween! 👻 I'm sharing rePossess 0.1.0, a Mac menu bar app that connects your reMarkable to Claude and other MCP clients: repossess.app
I picked up a reMarkable on a friend's advice. It's become my favorite way to take notes in grad school, but getting things on and off the tablet had me wishing for something more direct than its Google Drive integration. Something like "send this reading to my reMarkable" or "what did I write in last week's lecture?"
rePossess keeps a local copy of your library, refreshed over USB when the tablet is plugged in or through reMarkable's cloud after pairing. Agents can search typed text and PDFs, read typed pages as Markdown, and look at page images, handwriting included. Markdown goes back as a native notebook with editable typed text. PDFs and EPUBs go over as they are.
Want searchable handwriting? Turn on the optional recognition. It's off by default, and can use Qwen3-VL on your Mac through MLX (about a 5.8 GB download) or another provider.
Under the hood: a Swift port of rmscene to read and write the v6 page format, and cloud code that follows rmapi-js. Thanks to both projects for figuring out the hard parts.
This is a limited first-look for Apple silicon Macs running macOS 27 or later. It adds documents, with no edits, moves, or deletes. reMarkable's cloud API is unofficial, but has worked out so far on my Paper Pure.
It's free and open source (MIT).
Download it, if you dare 🎃
Today I'm sharing lore-swift, Swift bindings for Lore, Epic Games' new version control system: github.com/mattt/lore-swift
A few months ago, I started making a videogame with my kids. Until then, I'd led a semi-charmed existence of not having to think about Git LFS. But this asset-heavy project brought me back to the dark days of smudge filters and (apparently, given the price) gold-plated blob hosting. So after reading about Lore on HN, I decided to take a look.
I ran a simple trial on my Mac with the game's art, music, and sound effects so far: about 1,300 files and 395 MB, in Lore and in Git with LFS, side by side. After the import, Lore's local store took 4.8 MiB, against 375 MiB for .git (mostly the LFS cache). Five small edits to a 32 MB WAV file grew the Lore server by 1.9 MiB, and the LFS server by 155 MiB.
Provenance matters for this project, which mixes handmade and generated assets. Lore has built-in primitives for it, which is nice. Files, revisions, and branches can carry typed metadata that's part of immutable history, and files can declare tagged dependencies on each other. A generated sprite can point back to the drawing it came from, and the drawing knows what was made from it.
Lore isn't quite ready for my project yet. In the version I tried, locks didn't do much with auth off, and staging needed the server unless you cloned with a cache. I've also reported a couple of bugs so far (EpicGames/lore#263 & #264). But overall it's promising.
For the way we work, I wanted a native macOS app more than a command line, so I started with the foundation. lore-swift covers nearly all of Lore's C API with async/await, including progress streams for clone, push, sync, and interactive login.
These bindings are a start. I'll have more to share as I build on them!