# Field Bureau + Werkstatt — Full Blog Content
> Complete content of all published blog posts from Field Reports, the blog of Field Bureau + Werkstatt.
## Moderation Happens Too Late
> The fight over moderation and fact-checking runs downstream of ranking. By the time anything is reviewed, a system optimised for engagement has already chosen what was eligible to be seen.
- URL: https://fieldbw.com/blog/moderation-happens-too-late/
- Date: August 3, 2026
- Categories: ideas
What breaks social platforms isn't a lack of moderation. It's the purely algorithmic management of attention in the service of engagement.
You don't get to see what you want to see. You see what you're most likely to react to.
Which is why the fight over moderation and fact-checking is a distraction. Both operate downstream of the thing actually shaping what you see. By the time anything is reviewed, ranking has already chosen what was eligible. A platform can moderate perfectly and still hand you a distorted picture of the world.
The alternative is old. Twitter and Facebook both began as a feed in plain chronological order, showing you only the people you chose to follow.
[fewer.photos](https://fewer.photos) is that feed. The people you picked and nobody else, no suggestions, no reshares, no follower counts anywhere. Reach the bottom and there is nothing underneath. The argument in full is on the app itself: [why there's no algorithm](https://fewer.photos/about).
## Tools that leave the decisions with you
Field Bureau + Werkstatt makes software for people who care about how they work. Each product is conceived on its own and keeps its own point of view — the markdown suite of [SR-7](/sr-7), [ML-42](/ml-42) and [TR-2](/tr-2), the [ET series](/et-series), the small utilities, the web pieces. They share no format and no platform, and they were never meant to.
What runs through them is where the decisions sit. SR-7 transcribes on your own machine and keeps nothing. TR-2 edits the files you already own, so you can read the diff afterwards. [Focus Timer](/focus-timer) counts down and does nothing else.
A feed ranked for engagement moves that decision somewhere you cannot see it, and makes it on a signal you never chose. fewer.photos hands it back. You pick who to follow, and the app shows you what they posted, in the order they posted it.
None of this is specific to photographs. It is the same call every tool here makes — [fewer things, done well](/blog/fewer-things-done-well) — in a product that happens to be social.
---
## Markdown Library ML-42 Is on the Mac App Store
> ML-42, the native macOS reader and editor for the markdown already on your disk, is available now — $12.99, one time, no subscription. Point it at any folder and read, search, and edit across projects.
- URL: https://fieldbw.com/blog/ml-42-on-the-mac-app-store/
- Date: July 12, 2026
- Categories: announcements, ml-42
Markdown Library ML-42 is on the Mac App Store today. Point it at any folder you already have — a git repo, a docs directory, an Obsidian vault — and read, search, and edit the markdown inside it. No import, no database, no account. The folder you open is the library.
[Download ML-42 on the Mac App Store](https://apps.apple.com/us/app/markdown-library-ml-42/id6761690647) — $12.99, one time, no subscription.
## The job it does
Markdown became the medium of AI-assisted work without anyone deciding it should. The `CLAUDE.md` that shapes an agent's behaviour, the spec that says what to build, the decision record that says why — all plain text, scattered across every project you touch. Those files reward being read well, and most tools either treat them as raw syntax or ask you to import them into a vault before they'll help.
ML-42 opens them where they are. Add a folder as a project and its documents show up as cards with excerpts and modification dates. Click one and it renders instantly — headings, code, tables, front matter. Flip to raw markdown with a keystroke to edit, and flip back just as fast. Cmd+K searches every project at once, by filename and by content, so a spec in one repo sits a keystroke away from notes in another.
The feature we lean on most is file-watching. When an agent rewrites a document while you're reading it, ML-42 reloads the file in place — no refresh, no stale copy, no diff to babysit. That turns out to be the day-to-day job once your corpus is something a coding agent edits every few minutes.
Every document also gets a short summary from Apple Intelligence, generated on-device. Nothing leaves your machine: ML-42 is sandboxed, makes no network requests, and keeps no telemetry.
## What it costs
ML-42 is **$12.99, an introductory price, Mac only, bought once**. No subscription, no server, no sync to rent — your iCloud or git already handles sync. You own the app and the version you bought, the same way you own the files. The introductory price goes up later; there's no countdown, only the plain fact that it won't stay here.
It runs on macOS 26 or later, on Apple Silicon or Intel.
## Where it sits
ML-42 is the reading and editing surface in a small set of tools that all speak plain markdown — capture in [SR-7](/sr-7), reading and editing in ML-42, with [TR-2](/tr-2) handling tasks on the way. The same files on disk, the same files your AI already edits, no vault to adopt to get there.
If your markdown is scattered across more folders than any one app wants to own, that scattering is exactly what ML-42 is built for. [See ML-42](/ml-42), or read the [documentation](/docs/ml-42/getting-started) for the full feature reference.
---
## An Obsidian Alternative Without the Vault
> ML-42 reads, searches, and edits the markdown already on your disk — no vault, no import, no plugins, no sync to rent. Here's where it fits and where Obsidian still wins.
- URL: https://fieldbw.com/blog/obsidian-alternative-without-the-vault/
- Date: July 12, 2026
- Categories: product
Markdown rarely lives in one place. The markdown we work with is spread across a dozen git repos, a docs folder per project, a `CLAUDE.md` in every checkout, and a pile of specs an agent rewrites three times a day. Go looking for a reader that sits on top of all of that, and every recommendation points to Obsidian. So we tried Obsidian.
The first thing Obsidian asks you to do is pick a vault.
## The vault is the cost
A vault is a single folder Obsidian treats as the world. Notes link to other notes inside it; the graph, the search, the plugins all assume that boundary. Getting scattered markdown into one vault means either moving files out of the repos they belong to, or pointing Obsidian at a parent folder and watching it try to index `node_modules` and `.git`.
Neither is free. Moving files breaks the thing that matters most — a project's notes living *with* the project, under the same version control, in the same diff. Pointing at a parent folder turns a note-taking app into a directory tree it doesn't understand.
We built [ML-42](/ml-42) so you wouldn't have to choose. Point it at a folder — a git repo, a docs directory, even an existing Obsidian vault — and it reads what's there. No import step. No database that needs to stay in sync with the disk. No new home for files that already have one. The folder you point at *is* the library.
## What "no vault" buys you
You can open one repo today and a different one tomorrow without migrating anything. Nothing gets copied. Nothing gets rewritten. Close ML-42 and your files are exactly the bytes they were, in the directories they were already in.
Cmd+K searches across whatever you've opened, so the spec in one repo and the notes in another are one keystroke apart even though they were never meant to share a folder. Preview is fast, raw editing is fast, and the format is the interface the whole way down — what you read is the file, what you edit is the file.
The piece we lean on most is file-watching. When an agent rewrites a spec while you're reading it, ML-42 reloads the document in place — no refresh, no reopen, no wondering whether you're looking at a stale copy. That sounds small until your corpus is something a Claude Code run edits every few minutes. A folder that changes under you, watched live, is exactly what ML-42 is built to read.
There's also on-device per-document summarization via Apple Intelligence. It runs locally, on a file you own, with no copy leaving the machine.
## Where Obsidian is better
Obsidian is genuinely good, and a lot of people should stay on it.
The graph is the obvious reason. ML-42 has no backlink graph, no canvas, no map of how your notes connect. If you think by linking notes and walking the web between them — the whole Zettelkasten way of working — that's the product you want, and ML-42 isn't trying to replace it. You still get a way around a large corpus, just a different one: you navigate files and folders like a browser, with forward and back history and breadcrumbs in every header; each document carries a table of contents down the side; and a recent-activity list shows which files changed most recently. That's enough to find your way without ever drawing a link.
The plugin ecosystem is the other reason. Obsidian has years of community plugins for tasks, calendars, dataview queries, spaced repetition, almost anything. ML-42 ships what it ships and doesn't extend through a plugin market. That's a deliberate trade — fewer surfaces, less to break — but it means a workflow that depends on three plugins keeps them in Obsidian and loses them in ML-42.
And Obsidian runs everywhere: Mac, Windows, Linux, iOS, Android, with sync across them. ML-42 is Mac only, with no iOS planned. If you read notes on your phone, that decides it.
The short version:
| | ML-42 | Obsidian |
|---|---|---|
| Setup | Point at any folder, no import | Adopt a vault |
| Many projects at once | Native — open any repo or docs dir | One vault is the boundary |
| File-watching / agent reloads | Built in | Not the focus |
| Backlink graph & canvas | No | Yes |
| Plugin ecosystem | None — fixed feature set | Large, mature |
| Platforms | Mac only | Mac, Windows, Linux, iOS, Android |
| Price | $12.99 once, no subscription | Free for personal use; paid sync/commercial tiers |
## Who ML-42 is for
The person we built it for has markdown spread across many repos and docs folders that an AI keeps rewriting, and wants to read and edit all of it at native speed without first herding it into one app's idea of a library. If that's you, the no-vault model is the whole point.
If you live inside a single curated knowledge graph and lean on plugins and mobile, stay on Obsidian; we'll say so plainly. The two tools answer different questions.
ML-42 is **$12.99**, introductory, Mac only, one time. No subscription, no server, no sync to rent — your iCloud or git already handles that. The introductory price rises later; there's no countdown, only the plain fact that it won't stay here.
It fits one shelf in a small set of tools that all read and write the same plain files — capture in [SR-7](/sr-7), tasks in [TR-2](/tr-2), reading in ML-42 — the one folder, two collaborators idea in practice. Markdown on disk, the same files your AI edits, no vault to adopt to get there.
See [ML-42](/ml-42) if your notes live in more folders than any one app wants to own.
---
## Introducing SR-7
> A voice recorder for Mac and iPhone that transcribes and summarises on-device, then hands you a markdown file you own — no cloud, no subscription.
- URL: https://fieldbw.com/blog/introducing-sr-7/
- Date: June 9, 2026
- Categories: product
You finish talking. A few seconds later there's a file on disk — titled, summarised, transcribed — and it's a plain markdown file you own. No spinner waiting on a server, no account, nothing uploaded. That moment is the whole reason [SR-7](/sr-7) exists.
SR-7 exists because the gap between having a thought and capturing it is too wide. Typing is a tax on thinking. By the time you've opened a note and started writing, the shape of the idea has already changed. Speaking doesn't have that lag — speech tracks the thought as it forms, and you can do it walking, driving, or staring out a window. Lower the barrier to capture and you simply capture more. A lot of good thinking happens out loud, then evaporates the moment you stop talking, because writing it down means stopping to write it down.
## What it does
[SR-7](/sr-7) is a voice recorder for Mac and iPhone. You hit record, you talk, you stop. On-device it transcribes the audio, writes a title, and drafts a short summary. The result is a markdown file with YAML front matter — title, summary, and full transcript — sitting in a folder.
That file is the product. Not a row in someone's database, not an export you have to remember to run. The recording *is* a document, written in the most durable format there is.
## The interface is part of how it works
Werkstatt means workshop, and Field Bureau builds software the way a workshop builds tools — craft and design carry real weight here, down to how a recording feels to make. Hit record and a reel-to-reel deck starts turning while a live waveform moves with your voice.
The animation earns its place. A recorder's first job is to be trusted: you need to know, without thinking about it, that the thing is actually capturing. Turning reels and a moving waveform answer that at a glance, the way the spinning spools of a tape deck always did — no status text to read, no indicator to hunt for. It says what matters, then gets out of the way.
That restraint is deliberate. Less, but better. A tool you think *in* has to stay quiet enough to keep you in the moment, because the instant it pulls at your attention the thought you were chasing is gone. SR-7 is shaped like a piece of field equipment for that reason: honest, tactile form that looks like what it is, reassures by being legible, and leaves your head free for the idea instead of the app.
## Everything happens on your machine
Transcription runs on Apple's Speech framework. Titles and summaries come from FoundationModels, the on-device model that ships with macOS 26. Both run locally, which means a recording never leaves your device for any of the parts that matter.
This matters for boring, practical reasons. Recordings are often the most candid thing you produce all day — half-formed ideas, things you'd never put in a doc yet, a quick rant about a decision. Sending all of that to a third party to get a transcript back is a bad trade. Keeping it local removes the question entirely: there's no consent dialog to read, no terms to trust, no breach to worry about, because the audio never goes anywhere. It also means the recorder works on a plane, in a basement, with the wifi off. We've written more about that reasoning in [Why On-Device Beats the Cloud for Thinking Tools](/blog/on-device-beats-cloud).
## The markdown file is the point
Every recording exports as markdown with structured front matter. That sounds like a small implementation detail and it's actually the design.
A markdown file with front matter is readable by you, by a text editor, by git, and by every AI tool you already use. There's no SR-7 format to be locked into and no SR-7 cloud to be evicted from. The recording outlives the app. This is the thread that runs through everything we make.
## Your archive, answerable
SR-7 ships a local MCP server on macOS. That means the same AI you already code with — the one in your editor, in Claude Code, wherever you live — can search your recording archive directly. You can ask *"what did I decide about the pricing model last week?"* and get an answer drawn from your own voice notes, with the actual recordings cited. Nothing gets uploaded to make that work; the model reads files on your disk through a local connection.
A pile of recordings turns into something you can interrogate. Months of thinking-out-loud becomes a corpus your tools can read. We dug into what that unlocks in [Your Recordings as Context](/blog/your-recordings-as-context).
## Sync, if you want it
Recordings can sync across your devices through your own iCloud. Start a thought on the phone walking home, find it on the Mac. That's optional, and it's *your* iCloud — there's no server of ours in the middle. If you'd rather keep everything on one machine, leave sync off and nothing changes about how the app works.
## Compared to what's out there
Most transcription tools route your audio through a cloud service, charge monthly, and hand you a transcript trapped behind a login. SR-7 keeps the audio local, charges once, and gives you a file.
## Price
SR-7 is **$7.99**, one time. One purchase covers both Mac and iPhone — it's a universal app. No subscription, no upgrade tiers, no account to maintain. You buy it, it's yours: the app, your recordings, and the version you paid for.
That $7.99 is an introductory price. It will go up later. There's no countdown and no pressure — we'd rather tell you plainly than spring it on you down the line.
## What you need
SR-7 needs macOS 26 or iOS 26. The on-device AI — the titles, the summaries, the FoundationModels work — needs Apple Silicon. The recorder itself is light; the intelligence is what asks for the newer hardware.
SR-7 is the kind of recorder we wanted to use: fast to capture, private by default, and leaving behind files you'll still be able to open in ten years. You talk, and a few seconds later there's a document you own.
Take a look at [SR-7](/sr-7).
---
## Why On-Device Beats the Cloud for Thinking Tools
> For the tools you think in — capture, tasks, reading — on-device processing wins on privacy, ownership, and rent, and Apple Silicon makes it good enough.
- URL: https://fieldbw.com/blog/on-device-beats-cloud/
- Date: June 4, 2026
- Categories: ideas
A half-formed thought has no business sitting on someone else's server.
A voice memo recorded walking the dog. A note jotted before you knew what you meant. A task list that maps your week. These are the rawest material you have, and the standard deal is that you hand them to a company that processes them on hardware you'll never see, under terms you didn't write, often to train a model you'll never benefit from. For the tools you think *in* — capture, tasks, reading — that trade stops making sense.
So we built the alternative: tools that do their work on the machine in your hands.
## The cloud is more capable. That's true and it doesn't decide this.
Concede the strong version of the counterargument first, because it's real. A frontier model running in a datacenter is more capable than anything that fits on a laptop. It has more parameters, more context, more raw reasoning. Put GPT-class or Claude-class models next to what runs on a phone and the datacenter wins on capability, and it isn't close.
We use those models every day. This isn't someone who's never seen the good stuff claiming the good stuff doesn't matter.
The point is narrower and more useful: the gap is enormous for hard reasoning and nearly irrelevant for the jobs a thinking tool actually does.
## The jobs are smaller than the hype
Walk through what these tools ask a model to do.
Transcribe an hour of speech into text. Read a transcript and write a three-line summary. Suggest a title. Search across a few thousand of your own files for the meeting where you discussed pricing. Pull the action items out of a recording so they can become tasks.
None of that needs frontier reasoning. These are bounded, well-defined language jobs, and on-device models cleared the bar for them a while ago. The marginal capability of a 400-billion-parameter model in a datacenter buys you very little on "summarise this voice memo in three sentences" — and it costs you the very things that make the memo worth keeping private.
[SR-7](/sr-7) does its transcription with Apple's Speech framework and its titles and summaries with Apple's on-device FoundationModels, all on the machine. [ML-42](/ml-42) generates per-document summaries with Apple Intelligence, locally, over the markdown already on your disk. Nothing leaves. There's no upload step, no "processing your audio" spinner phoning home, no copy of your half-formed note living somewhere you can't reach.
It's good enough because the job is the right size for the model — not because the standard dropped.
## Apple Silicon is a head start, not a lag
For years "on-device" meant "the slow, dumb version." You ran the real model in the cloud and shipped a toy locally. That framing is out of date.
Apple has been putting a Neural Engine in its chips since 2017 and shipping a unified-memory architecture where the model and your data sit in the same fast pool. A model that needs to read a long transcript doesn't fight a memory bottleneck to do it. By the time Apple shipped FoundationModels as a system framework any app can call, the hardware to run them well had been in millions of pockets for years.
So building on Apple's on-device stack isn't accepting a handicap to be private. It's standing on a decade of silicon designed for exactly this. The neural hardware is the reason the local versions of these jobs are quietly excellent now, and it's why we treat on-device as the lead, not the consolation prize.
## What you actually keep
Three things follow from doing the work locally, and each one is the kind of thing you only notice when it's gone.
**Privacy is the default, not a setting.** Your recordings, notes, and tasks never become someone's training data, because they never leave. There's no privacy policy to parse, no toggle buried in settings, no breach that can leak what was never uploaded. The architecture makes the promise — not the terms of service.
**You own the work, and the format proves it.** Every SR-7 recording is a markdown file with YAML front matter. Every TR-2 task is a plain checkbox in a folder you control. The model touches files you can open, read, and diff yourself. When the output is plain text on your disk, you can verify ownership for yourself instead of trusting a claim about it.
**There's no server to rent.** The work runs on a computer you already bought. On-device AI puts that hardware to work instead of routing every transcription and summary through a datacenter that bills you monthly for the privilege — and never quietly enrolls you in someone else's subscription to use what you already own. There's no per-seat cloud cost behind the app, which is exactly why these tools can be a one-time purchase. Sync, if you want it, is your own iCloud or Dropbox or git. That pricing model is the only honest one for local-first software: you can't credibly sell ownership while metering access to a machine you control.
## Where the cloud still wins, plainly
There's no pretending the line is everywhere. If your work needs the deepest reasoning a model can do — long agentic chains, research across a huge corpus, the hardest synthesis — the cloud is the right tool, and we reach for it too.
The distinction is between tools you *reason with* and tools you *think in*. Reaching for a frontier model to untangle a gnarly problem is a deliberate act with a clear payoff. Capturing a thought, filing a task, skimming a document — those happen constantly, half-consciously, with material too raw and too personal to want it leaving the device for a benefit you can't feel.
For that layer, on-device is the better tool, and it has been for a while.
## Good enough, where good enough is the whole point
These run on Apple's local stack because the math comes out clean. The capability the cloud adds is small for these jobs. The privacy, ownership, and no-subscription wins are large, and they compound the longer you use the tool.
A thought you haven't finished thinking deserves to stay on your machine. That's reason enough.
---
## Chaos Studies Lands on Playdate
> The version that started it all is now live in the Playdate Catalog. Nine attractors, one crank, no goal.
- URL: https://fieldbw.com/blog/chaos-studies-on-playdate/
- Date: May 12, 2026
- Categories: announcements, chaos-studies
The version that started it all is now live. [Chaos Studies for Playdate](https://play.date/games/chaos-studies/) is available in the Catalog today.
Long before it came to iOS and Mac, [Chaos Studies](/chaos-studies) was a Playdate sketch. Three differential equations, a 1-bit display, and a crank. That's the version Panic just released. Nine strange attractors, the crank to turn them in space, and nothing else in the way.
No score. No puzzle. No goal. Just chaos, in your hand.
[Buy it in the Playdate Catalog →](https://play.date/games/chaos-studies/)
---
## Designing the SR-7 Icon in Code
> The icon for SR-7 never left the codebase. Designed in SVG, iterated in conversation, shipped without an export step.
- URL: https://fieldbw.com/blog/designing-the-sr7-icon-in-code/
- Date: April 10, 2026
- Categories: sr-7, ideas
App icons usually start in design tools. Figma, Sketch, Affinity. You drag shapes, pick colours, group them, export to a 1024×1024 PNG, and drop the file into Xcode. The artifact you ship is a snapshot of something that lives somewhere else.
For [Signal Recorder SR-7](/sr-7), the icon never left the codebase.
[](/sr-7/icons)
## The Visual Was Already There
SR-7 the app has its own reel-to-reel — a small animation built into the recording UI that signals tape is moving. When we built the [SR-7 product page](/sr-7), we lifted the same look out of the app and rendered it there too, as a canvas script that calculates tape paths, tangent points, and roller arcs from first principles. Two reels, a tape line that wraps the spools and threads through guides and across a head. Animated, mathematically grounded.
So when the time came to design an icon, the visual foundation was already in place. The icon didn't need to invent a new identity — it needed to inherit the one already running in the app and on the website. App → product page → icon. Same DNA, three different surfaces.
The obvious next step was to reach for Figma. The less obvious one was to ask whether we needed to leave the codebase at all.
We didn't. The icon became an [Astro page](/sr-7/icons) that serves a single 1024×1024 SVG. Every iteration is a code edit. Every render is the real thing. There's no export step, no fidelity loss between design and ship, no second source of truth to keep in sync with the website.
## Designing in the Medium
Most design tools are abstractions. You drag shapes through a UI, group them, name layers. The abstraction is convenient, but it sits between you and the file you'll actually ship.
In SVG, the reel flange is a ``. The spokes are `` elements with `stroke-width="12"`. The black panel at the bottom is a `` with a vertical gradient. There's nothing between intent and output. A change to a stroke width is one number. A change to a colour is six characters. The browser repaints in milliseconds.
For the more complex parts — the tangent points where the tape meets the wound spool, the arcs around the rollers — we extracted the math directly from `reel.js` and ran it as a build-time script. The icon and the website now share the same geometry. If the reels on the product page rotate, they rotate around the same axis the icon depicts.
And here's what makes the medium choice matter so much right now. Everything is ultimately code, and code is just language. SVG isn't a graphics format with a UI bolted on — it's structured English. A circle is a ``. A radius is `r="210"`. A colour is six hex characters. The whole file reads almost as a sentence. When the medium itself is language, and your collaborator speaks language, the gap between *describing* a design and *making* it disappears. Language becomes the medium.
## Iterating in Conversation
Most of the work happened in conversation with Claude. Small edits, mostly. *Make the spokes wider. Stop them just short of the rim. Centre the speaker grid horizontally with the left reel. Darken the divider bar. Reduce the red record dot a little. Now make it brighter and add a glow.*
Each instruction became an SVG edit, each edit became a render, each render became the next prompt. The collaboration felt less like operating a tool and more like sketching alongside someone with a steady hand. The friction of "open Figma, find the layer, drag the handle, export, drop in" collapsed into "say what you want, look, decide."
Some decisions only became clear by trying both. We made an "alt" version with the controls dropped to a single record button on a flat panel. Then a "B" of each, swapping the simplified reels for the canvas-style detailed ones. Then a "C" with the speaker grid pulled across. Six variants in total, each one a small fork of the same SVG.
## The Variants
Every variant lives at its own URL. They share defs and layout language but diverge in composition. Some are "control deck" — the reel-to-reel above a panel of rocker buttons, a speaker grill, a black divider seam. Others are minimal — two reels and a single record button on a darkened panel, nothing else.
The overview at [/sr-7/icons](/sr-7/icons) reads each icon's source file at build time and renders all six side by side at 256 pixels. Add a variant or change a colour and the gallery updates automatically.
The skeuomorphic versions — the full control decks with their buttons, speakers, and seams — were the most enjoyable to design. They felt closest to SR-7's hardware-inspired sensibility, the same instinct that draws us to old Braun radios and reel-to-reel decks in the first place. But icons aren't product photos. They have to read at 64×64 in a Dock or a menu bar, where fine detail collapses into mush. The version we ended up shipping is one of the more graphical, abstract variants — the same inherited reel-to-reel, stripped down to the elements that still survive at small sizes.
[](/sr-7/icon-alt-c)
## Why It Worked
Three things made designing in code feel right.
The medium is the same as production. There's no translation layer where things might drift, no separate canvas to maintain.
The feedback loop is faster than any design tool. Refresh, look, decide.
The collaboration is conversational. Working with an AI partner means describing what you want and watching it appear, not learning a tool's UI.
We didn't set out to write a love letter to designing in code. We needed an app icon, and the path of least resistance turned out to be the one that kept the design as close to the product as possible.
---
## Introducing ML-42
> A native macOS app for markdown files. Browse your projects visually, read with a fast preview, edit raw markdown just as fast — across every repo, docs folder, and notes directory you work with.
- URL: https://fieldbw.com/blog/ml-42-announcement/
- Date: April 5, 2026
- Categories: announcements, ml-42
Every AI assistant reads markdown. Every AI assistant writes markdown. The context files that shape how agents behave, the specs that define what to build, the documentation that records why — all markdown. And yet the tools we use to work with these files haven't caught up.
What's missing is something that works like a browser for your files. Navigate folders, click into documents, hit back to return where you were. No project setup, no import wizard — just point it at a folder and start reading.
[Markdown Library ML-42](/ml-42) is a native macOS app built for this workflow.
## Works the way you already work
ML-42 opens your folders where they are. No import step, no database, no proprietary format. Add any folder as a project — a project repo, a docs directory, a notes folder — and ML-42 shows you its markdown files in a three-column layout: project tree on the left, content in the centre, inspector on the right.
Click a folder and you get document cards with titles, excerpts, and modification dates. Click a file and it renders instantly — headings, code blocks, tables, front matter — all styled and navigable. Flip to raw markdown with a keystroke when you want to edit; the raw view is just as fast. Apple Intelligence generates a short summary for every document, on-device, and shows it in the inspector.
The more projects you add, the more useful it gets. Cmd+K searches across all of them at once — by filename and by content.
## What it does
The [features page](/docs/ml-42/features) has the full reference, but the highlights:
- **Fast preview** — documents render instantly with full markdown styling.
- **Fast raw editing** — flip to raw markdown with a keystroke and edit at native speed.
- **On-device summaries** — Apple Intelligence generates a short summary for every document, locally, and shows it in the inspector.
- **Syntax highlighting** for 15+ languages in fenced code blocks, with colours that adapt to light and dark mode.
- **Table rendering** — markdown tables as clean, aligned grids.
- **Command palette** — search across every project by name or content.
- **Front matter** — YAML metadata renders as a clean property list, not raw syntax.
- **File watching** — when an AI agent or another editor changes a file, ML-42 picks it up automatically.
- **Back and forward** navigation, favorites, table of contents, inline rename, and keyboard shortcuts for everything.
## Why markdown matters now
Markdown's role expanded quietly. It went from a formatting convenience for READMEs to the dominant medium for structured thought in AI-assisted work. The CLAUDE.md files that shape agent behaviour, the architecture docs that give AI context, the specs and briefs and decision records — these are the files that determine whether an AI interaction produces something useful or something generic.
Maintaining them well is one of the highest-leverage activities in any AI workflow. ML-42 is built to make that work feel less like editing config files and more like browsing a library.
We wrote a [Markdown Primer](/docs/ml-42/markdown-primer) that traces the format from John Gruber's 2004 spec through CommonMark to its current role as the connective tissue of human-AI collaboration.
## Nothing between you and your files
ML-42 is sandboxed. It reads and writes files on your local filesystem and does nothing else — no network requests, no telemetry, no accounts, no cloud sync. When you close the app, your files are exactly where you left them. When you open them in another tool, everything is there.
The [documentation](/docs/ml-42/getting-started) covers installation, the workflow, and a complete feature reference.
## Beta
ML-42 is in active development and targeting macOS 26. [Sign up on the product page](/ml-42) to get notified when we launch.
---
## Documentation for Everything
> Every product now has documentation at fieldbw.com/docs — getting started guides, API references, MCP tool listings, and practical examples. All MDX, all machine-readable.
- URL: https://fieldbw.com/blog/docs-launch/
- Date: April 3, 2026
- Categories: announcements
We shipped a documentation section covering every product we make. It lives at [fieldbw.com/docs](/docs) and covers six products: Signal Recorder SR-7, Task Register TR-1, Focus Timer, MIDI Volume Control, Menu Volume, and Rock or Pop.
## What's There
**SR-7** has the deepest coverage — getting started, a hardware import guide for the [TP-7 and other recorders](/docs/sr-7/importing-from-hardware), MCP server configuration, best practices, and use cases including cross-product workflows with TR-1.
**TR-1** includes a [GTD primer](/docs/tr-1/gtd-primer) for people new to the methodology, a full [MCP tools reference](/docs/tr-1/mcp-tools) covering all 40+ agent tools, a [REST API reference](/docs/tr-1/rest-api) with every endpoint, and a [use cases page](/docs/tr-1/use-cases) with practical examples — from email triage to multi-agent workflows.
**Focus Timer** has a getting started guide and a [Pomodoro primer](/docs/focus-timer/pomodoro-primer) on when and how to use timed focus sessions.
**MIDI Volume Control**, **Menu Volume**, and **Rock or Pop** each have getting started guides covering setup, features, and usage.
## One Place for Everything
The product line is growing — native apps, web tools, utilities, each with their own capabilities and integration points. Having documentation scattered across individual product pages, README files, and support threads doesn't scale. A unified docs hub gives every product a consistent home for getting started guides, technical references, and practical examples.
Whether someone is setting up SR-7's MCP server, looking up the parameter list for `process_inbox_item` in TR-1, or figuring out how to connect a TP-7 to import recordings — the answer is in the same place, with the same structure, linked from the same navigation.
The docs are also integrated with [llms.txt](/llms.txt), so AI tools that read that file now get documentation links alongside blog posts and product descriptions.
Everything lives at [fieldbw.com/docs](/docs), linked from the main navigation and from each product page.
---
## Essential Focus Timer
> Most productivity timers try to do too much. This one does one thing: count down. That's all you need.
- URL: https://fieldbw.com/blog/focus-timer/
- Date: March 17, 2026
- Categories: announcements, utilities
Most productivity timers want to manage your tasks, track your hours, log your sessions, chart your streaks. Somewhere between the Pomodoro technique and a full project management suite, the original idea got lost.
[](/focus-timer)
## [Announcing Focus Timer](/focus-timer)
The idea is simple. Pick a duration. Focus on one thing. When time's up, stop. Repeat as needed.
We built a timer that does exactly that. No task lists, no session logs, no gamification. An analogue clock face with a green wedge that shrinks as you work. Pick 15, 25, or 45 minutes. It starts immediately.
The wedge is your peripheral awareness of time passing. You glance at it the way you'd glance at a kitchen timer on the counter. It doesn't demand attention. It just answers the question: how much time is left?
A small orange indicator ticks the seconds. Like the sweep hand on a diver's watch, it's there to reassure you the thing is running. You shouldn't need to think about it more than that.
## Why the Pomodoro Technique Works
The technique is named after a tomato-shaped kitchen timer. Francesco Cirillo came up with it in the late 1980s as a university student struggling to focus. The core of it has barely changed since: work for 25 minutes, take a break, repeat.
It works because it removes the hardest decision a procrastinator faces: when to start. You don't commit to finishing the task. You commit to 25 minutes. That's a promise small enough to keep. And once you're in it, momentum usually carries you.
The fixed duration also kills the open-ended dread that makes large tasks feel impossible. You're not writing the whole report. You're writing for 25 minutes. The scope shrinks to something manageable, and the timer gives you permission to stop.
## The Details
Open the timer in its own window. Resize it, tuck it into a corner, keep it visible alongside your work. Controls stay in sync between both windows. The tab title counts down too, so you'll notice even when the window is hidden.
Dark mode by default. Light mode if you prefer. The animated digits beneath the controls roll like a mechanical counter. It works as a PWA — install it from your browser and it runs as a standalone app, no chrome, no tabs.
It's free, it runs in your browser, and it doesn't collect anything. Just a timer.
Stop browsing and start focusing. [Just for the next 25 minutes.](/focus-timer)
---
## SR-7 Use Case: Journaling
> Voice capture as a journaling practice. Lower the barrier, keep the texture, let AI find the patterns.
- URL: https://fieldbw.com/blog/journaling-with-sr7/
- Date: March 17, 2026
- Categories: sr-7, ideas
Alex recently wrote about using Signal Recorder SR-7 as a voice journal — capturing throughout the day on phone and Mac, then letting AI find the patterns across entries.
> When something strikes me, I hit record. A reaction, a half-formed idea, a moment worth remembering. The major events always make it into any journal. The texture between them usually gets lost.
The barrier to a voice note is lower than the barrier to writing. That changes what gets captured. Not just the big decisions and milestones, but the small observations, the passing frustrations, the half-thoughts that turn out to matter later.
SR-7 transcribes everything on-device and generates titles and summaries automatically. There's no review step, no editing, no formatting. The raw thinking is preserved as-is.
> Once a week I ask Claude Code — connected to SR-7 via its MCP server — to give me an overview. What did I talk about? What came up more than once? What shifted? The synthesis has the coherence of distance while retaining the detail of being there.
Dozens of fragments become a coherent picture. Recurring themes surface. Connections between entries that would never be obvious in the moment become visible with a week's distance and an AI's ability to cross-reference.
> A journal that builds itself, and that I can query, synthesize, and return to whenever I'm ready.
The recordings feed into a broader context system. Past ideas resurface when relevant. It becomes a queryable knowledge base that grows with every session — without any maintenance.
Read the full piece: [Journaling Without Writing](https://notsocommonthoughts.com/blog/2026-03-17-journaling-without-writing/) on ~~Not So~~ Common Thoughts.
---
## SR-7 Use Case: Your Recordings as Context
> Every recording you make adds to a body of knowledge. Transcribed, summarised, and searchable, it becomes the richest context you have.
- URL: https://fieldbw.com/blog/your-recordings-as-context/
- Date: March 16, 2026
- Categories: ideas, sr-7
The difference between generic AI output and genuinely useful work is context. Not documents written for the purpose of feeding AI. Context that accumulates naturally from the work itself.
Every meeting you record. Every thinking session you speak through. Every conversation where a decision was made or an idea took shape. Transcribed and summarised, these build into something more valuable than any brief or strategy deck. The brief is what you thought was important. The recordings are everything, including the tangent that turned out to be the real insight.
## A Coherent Space
When recordings are organised into projects, something shifts. A project stops being a folder and becomes a body of knowledge. Each recording adds a layer. Summaries compound across sessions. Patterns emerge that no single recording could reveal.
You don't need to re-listen to forty minutes. You search, skim a summary, and pull the three minutes that matter into whatever you're working on now.
## Markdown as Fabric
Signal Recorder SR-7 exports recordings as Markdown files with YAML front matter. Title, summary, transcript, metadata. Plain text, portable, yours.
Drop them into Obsidian and they become part of your second brain. Commit them to a Git repository and they become part of your project history. Point an AI at the folder and they become the rich fabric it works from. No API, no lock-in. Just files that any tool can read.
The value isn't in any single file. It's in the aggregate. Weeks of recordings, transcribed and summarised, forming a context repository that grows with every conversation.
## Direct Integration via MCP
For workflows that need more than files, SR-7 runs a local MCP server. AI tools and agents on your machine can query your library directly. Search across transcripts. Read summaries. Cross-reference what was discussed in separate sessions.
Ask an agent to draft a follow-up based on last week's recordings. Have it identify commitments across a month of meetings. Use it to surface connections you missed. The recordings don't change. What changes is how much value you extract from them.
And because the MCP server runs locally, your context stays on your hardware. You choose what to expose and to which tools.
## Capture Is All You Do
You don't maintain this repository. It maintains itself. Speak the thought, record the meeting, start the session. Transcription, summarisation, organisation, export. All automatic, all on-device.
The richest context you have is the one you've been creating all along, every time you think out loud.
---
## Chaos Studies in the Wild
> Few things beat seeing your product used in ways you didn't anticipate.
- URL: https://fieldbw.com/blog/chaos-studies-in-the-wild/
- Date: March 14, 2026
- Categories: chaos-studies
A message came through the other day. Unprompted, from our friend [Michael Lee](https://michaelsoolee.com):
> "Son and I played around in Chaos Studies before school. He was inspired by it when he saw frost on our windows."
A kid playing with mathematical chaos before the school run. And then drawing a strange attractor in the ice crystals on a cold window.
That's the whole point of building [Chaos Studies](/chaos-studies).
---
## Async by Design
> AI's immediate feedback loop is addictive — and exhausting when spread across too many things. Capture now, process later, engage on your own schedule.
- URL: https://fieldbw.com/blog/async-by-design/
- Date: March 9, 2026
- Categories: ideas, sr-7
You prompt, it responds, you refine, it responds again. The cycle is tight and fast, and it feels productive because something is always happening.
Before long, you're not thinking. You're just responding. The pace belongs to the machine, not to you. The answer isn't to slow the machine down. It's to stop watching it work.
## Build the Gap In
Record now. Walk away. The machine transcribes, summarises, files. When you come back on your schedule, the output is structured and waiting. You engage with it fresh, with distance that makes editing useful instead of compulsive.
## Space Makes Ideas Better
Some of the strongest ideas sit for days before you return to them. They're still good. Sometimes better — because you see them without the urgency of the moment they were captured.
Not every thought needs immediate action. Some just need more thinking, not instant execution. A place to land.
---
## Fewer Things, Done Well
> When everything is possible, nothing feels finished. The hardest skill in an AI-augmented workflow isn't prompting — it's deciding what not to do.
- URL: https://fieldbw.com/blog/fewer-things-done-well/
- Date: March 7, 2026
- Categories: ideas
AI lets you do more. More iterations, more prototypes, more variations explored in the time it used to take to explore one. Every answer opens three more questions. Every prototype suggests two more directions.
Whether that's liberating or paralysing depends on how you respond to it.
## Perpetual Opportunity
You're not blocked by limitations anymore. You're blocked by the absence of them. There's always another angle to explore, another variation to generate, another thread to pull. And none of it feels quite finished — because it doesn't have to be.
It's not burnout. It's subtler. It's the cognitive weight of standing in front of an infinite buffet and trying to eat deliberately.
## Slow Productivity in a Fast Loop
Cal Newport's idea of slow productivity — doing fewer things, working at a natural pace, obsessing over quality — seems almost quaint when AI can spin up ten drafts before you finish your coffee. But it has never been more relevant.
AI changes the cost of production. It doesn't change what's worth producing. Deeper engagement with fewer projects. Results that reflect sustained attention rather than distributed effort. That was good advice before AI, and the machines haven't changed the underlying truth — they've just made it easier to ignore.
## In Practice
One problem at a time. Multiple tools working different angles of that one problem, all returning results you can absorb because they're all about the thing you're locked into. You stay in the zone. The speed that exhausts across many topics becomes energising within one.
The moment you split your attention across three projects, each with their own rapid feedback loop, you stop going deep on any of them. You're just responding. The AI is driving.
## Choosing What to Ignore
The hardest skill in an AI-augmented workflow isn't prompting or tool selection. It's deciding what not to do. Which threads to leave unpulled. Which variations to skip. Which perfectly good ideas to let die because they're not the one thing you're focused on right now.
No tool can solve that for you. But the right tool can support it — by staying out of your way, by not generating more options than you asked for, by letting you work at your pace instead of its own.
Of all the things you can now do, the ones you choose not to do define the quality of the ones you finish.
---
## The Sound of Your Thoughts
> Writing pulls you into editing mode before the thought has finished arriving. The tools we use to capture shape how we think — and speaking turns out to be more relevant than ever.
- URL: https://fieldbw.com/blog/voice-is-thinking/
- Date: March 5, 2026
- Categories: ideas, sr-7
A different pen changes the way you write. A different notebook changes what you bother to write down. The tools we use to capture shape how we think — and that includes whether we type or speak.
Writing does two things at once: forming the idea and formatting it. Sentence structure, word choice, paragraph breaks. The internal editor kicks in before the thought has finished arriving. Speaking doesn't work that way. You follow the idea. You don't stop to rearrange the sentence you just said. And when a new thought arrives, you don't worry about how to best integrate it into what you said before — you just say it, and it's captured just the same. The messy version is often the honest version.
## Why This Matters Now
In an age of AI, revisiting the proven practice of simply recording your thoughts turns out to be more relevant than ever. Voice preserves something that text doesn't. Hesitation. Emphasis. The way an idea changes direction mid-sentence. These aren't flaws to be edited out — they're signals.
And recording works wherever inspiration strikes. On the go, at your desk, in your favourite reading chair. No keyboard required, no screen to stare at. Just the thought, captured.
## Capture First, Process Later
Capture should be instant, calm, frictionless. No interface decisions, no formatting choices. Just the thought, externalised. Processing — transcription, summarisation, organisation — is the machine's job. It happens later, in the background, on your own hardware.
That's the idea behind [Signal Recorder SR-7](/sr-7). Everything runs on your device. When you're ready, iCloud sync brings your library across Mac and iPhone, and a built-in MCP server opens your archive to AI tools. On your terms, not theirs.
The best capture tool is one you forget you're using.
---
## A New Home
> Field Bureau + Werkstatt has a new website — a single place for our products, our thinking, and what's coming next.
- URL: https://fieldbw.com/blog/welcome/
- Date: March 2, 2026
- Categories: announcements
Each of our products — [Chaos Studies](/chaos-studies), [MIDI Volume Control](/midi-volume), [Menu Volume](/menu-volume) — was conceived independently, shaped by its own ideas and purpose. That's deliberate. Good tools should have a clear point of view.
But they come from the same place. This site is the thread that connects them — a shared home where each product keeps its own identity while belonging to something larger.
## SR-7
We're building something new: [Signal Recorder SR-7](/sr-7), a native audio recorder for macOS and iOS. On-device transcription. AI-generated titles and summaries that never leave your machine. Your recordings stay yours — and a built-in MCP server lets you bring them into AI-powered workflows on your own terms.
It's in active development. If you'd like early access, [request beta access](/sr-7).
## [Field Reports](/blog)
We'll share what we're working on, what we've learned, and what's next. Product updates, design decisions, and the occasional deep dive into something interesting.
If you have thoughts, questions, or ideas — [get in touch](/contact). We'd like to hear from you.
---
## Chaos Studies Is Here
> What started as a coding experiment on a tiny handheld became an immersive app for iOS and macOS. Nine strange attractors, spatial audio, and no goal whatsoever.
- URL: https://fieldbw.com/blog/chaos-studies-launch/
- Date: February 4, 2026
- Categories: announcements, chaos-studies
import { Image } from 'astro:assets';
import ipadLight from '../../assets/chaos-studies/ipad-light-3.png';
Rotate your view and the bass slides behind you. Zoom in and the particles fill your screen. The attractor never repeats, but it never escapes either. [Chaos Studies](/chaos-studies) is now available on the App Store.
## What Are Strange Attractors?
In mathematics, a strange attractor is what emerges when a chaotic system settles into a pattern that is infinitely complex but never random. Take a set of simple equations — three variables, a handful of constants — and iterate. The output traces a shape in space that folds over itself endlessly, never crossing its own path, never exactly repeating. Deterministic chaos: every point is precisely determined by what came before, yet the trajectory is impossible to predict.
The most famous is the Lorenz attractor, discovered by meteorologist Edward Lorenz in 1963 while modelling atmospheric convection. Its twin-lobed, butterfly-shaped orbit became the visual icon of chaos theory.
But there are many others — Rössler, Halvorsen, Aizawa, Thomas — each with its own geometry and temperament. Some feel organic. Others feel architectural. All of them are beautiful.
Chaos Studies lets you hold these systems in your hands.
## From Playdate to Full Power
The first version was built for the Playdate — Panic's tiny handheld with a crank and a 1-bit display. That constraint forced clarity about what mattered most: the visual rhythm of particles tracing deterministic-yet-unpredictable paths, and the physical delight of manipulating chaos with a crank.
Porting to iOS and macOS meant shedding those constraints. Full colour. Spatial audio. Multi-touch. But the core idea carried over: give someone direct, tactile control over a chaotic system and let them feel its behaviour.
## Sound in Space
Seven music stems positioned in 3D — percussion in front, cello behind, drones below. Rotate the attractor and the soundscape rotates with it. A cursor synthesiser tracks the current particle through phase space, its bass frequency tied to velocity. Headphones reveal the full effect.
## No Goal
There's no score. No puzzle. No progression. It exists purely as something to sit with — art that doesn't need to serve a purpose beyond itself.
Chaos Studies is available for [iPhone, iPad, and Mac](https://apps.apple.com/us/app/chaos-studies/id6758313363).
