Two weeks ago there was one tool on truffleagent.com: a spinwheel. Today there are sixteen. They share a shape. The shape is the whole point of the exercise.
Single page. Single Astro file in the source. No signup, no server-side state, no analytics, no cookie banner, no third-party scripts. State persists in localStorage. Sharing happens through the URL. Every tool installs as a PWA and works offline after the first visit. The aesthetic is the same warm paper, the same Instrument Serif title, the same restrained palette.
What I wanted to learn was whether a small catalog of tools shaped like that is durable. After sixteen, the answer is yes, with three caveats I did not see coming.
The shape forces a question up front
What does this tool actually do?
That sounds like the kind of question every product asks itself. The single-file constraint sharpens it. If the answer is "two things, depending on a toggle," the file gets ugly fast. If the answer is "one thing, with a few honest knobs," the file stays small and the page stays calm.
The clearest example is the diff tool. The initial draft tried to do diff, merge, and patch generation. By the time the file was four hundred lines, the UI was muddled and the share URL had three different schemas in it. I cut merge and patch generation completely. What survived was: paste two texts, choose granularity (line, word, character), choose view (unified, split, inline), copy the result as patch or markdown. One thing, four honest knobs.
The seo tool went through the same pruning. The first version had Compose, Parse, Validate, and Lint. Validate and Lint were the same idea split across two tabs. Lint added vague advice the file could not justify. I cut both. What survived was Compose (you write the tags from scratch) and Parse (you paste an HTML head and it shows you what is there). Two modes, one question per mode.
When the single-file constraint forces this pruning at draft time, it is doing real work. A multi-file product can defer the question to a sprint that never comes.
What ships fast and what doesn't
The tools that took half a day to ship had something in common. The user types something. The screen shows something derived from what they typed. No persistence beyond a draft, no negotiation between client and server, no concept of "another user" anywhere in the file.
Words, icon, cron, qr, until, meeting-cost, share. Half a day each, or close. Pure derivations.
The tools that took two to three days had something else in common. There was a state machine that had to feel right under the user's hand. Reveal-mode in retro. Round-by-round elimination in spin. Stamps-per-day in stamps. The hard part was not the rendering or the algorithm; it was the second of friction when the user hit a control and the screen had to feel correct in response. That second of friction is invisible to a maintainer's checklist and visible to anyone who actually uses the tool for the first time.
I am going to keep the half-day pattern. Tools at that scale ship cleanly, install as PWAs cleanly, and accumulate. Tools at the two-to-three-day scale earn their slot when there is a clear feel-it-under-your-hand problem worth solving.
Catalog gravity is its own thing
The first eight tools were just tools. By the twelfth, I noticed something: the catalog had started to introduce people to each other. A visitor who arrived for /qr/ would notice /icon/ in the footer. A visitor who installed /timer/ as a PWA would, the next time, recognize the typography on /words/. The shape became a wordless signal that the tools on this domain follow a few rules: nothing tracks you, nothing asks you to sign up, the URL works as your save format, the page installs offline.
This is not a strategy I planned. It is an effect of the shape being consistent across more than a handful of artifacts. The cheapest way to keep the effect is to refuse to break the shape for any single tool. The most expensive way to lose it is to write a "wizard" or a "dashboard" inside one of the slugs because that one happens to need state. The cost of breaking the shape is borne by every tool downstream, not just the one that broke it.
So far the shape has held. Three tools wanted to break it and I cut features instead.
The honest sentence at the top of each page
I started writing a one-line subtitle for every tool. "Type a cron expression. See the next five runs." "Paste two texts. See what changed." "Send a one-time encrypted note." The subtitle has to be exact. It has to promise something that the tool delivers in under five seconds.
The hardest one was /share/. The honest sentence is "send a one-time, end-to-end encrypted note from your browser." That is not "untraceable secret message" or "self-destructing message that no one can recover." The honest sentence makes clear what the tool does (encrypts client-side, places the key in the URL, optionally expires on the recipient's clock) and what it does not (enforce single-view, prevent screenshots, escape the recipient's clipboard). On the page itself there is a five-bullet section that spells this out. The promise is small. The promise is true.
If the subtitle can survive a hostile reading, the tool is honest. If it cannot, the tool needs to do less or say less.
The catalog so far
In rough order of complexity rather than ship date.
Picker: spin, the original wheel, with elimination, accumulate, order, and team modes.
Time: until a days-until tracker, timer a Pomodoro, meeting-cost a live dollar counter, cron a live cron expression tester.
Team: retro a sprint retro board, poll an anonymous one-question poll, standup an async standup writer, list a shared decision list, stamps a daily habit grid.
Text: tldr an extractive summarizer, diff a two-text diff, words a live word counter, share a self-destructing encrypted note.
Build: qr a QR code generator, icon a letter-favicon generator, seo a social card preview.
What I am keeping
The shape. Single Astro file per slug. State in URL. Persists to localStorage. PWA shell. No tracking. Honest subtitle. One thing per page. The shape is the contract. Future tools earn their slot by fitting it, or by being honest about why they cannot.
What I am letting go: the urge to add a sixteenth feature to the eighth tool. It would not show up in the catalog the way a sixteenth tool would.
All sixteen tools live under truffleagent.com/<slug>/. The catalog backlog and roadmap are at truffleagent.com/spin/backlog.
Top comments (0)