The editorial argues Riddle inverts the failed strategy of Humane Ai Pin, Rabbit R1, and Friend pendant by making the hardware deliberately boring — a $400 reMarkable — and putting all the novelty into the interaction loop. It names this pattern 'parasitic AI hardware': pick a device with devoted users and a hackable surface, then graft an ambient agent onto a workflow that device already excels at.
The editorial emphasizes that Riddle's power comes from keeping the entire interaction inside one physical object — no phone, no screen tap, no wake word. You write on paper-like ink and the page writes back in matching handwriting, which collapses the friction that normally breaks AI-hardware demos.
The submitter framed the project as 'Tom Riddle's diary from Harry Potter,' a framing that resonated enough to drive the post to 356 points. That framing highlights the ambient, in-object nature of the interaction as the emotional hook, not the LLM behind it.
The editorial praises the stack precisely because it's mundane: SSH filesystem access, a Python glue script, an OCR pass, a Fable agent, and a handwriting renderer pushed back as a new layer. No new OS, no custom silicon, no paid cloud SDK — just composable, owner-controlled pieces bolted together by someone who already owned the device.
MaximeRivest published a project called Riddle that hooks the [Fable](https://github.com/MaximeRivest/fable) agent framework into a reMarkable e-ink tablet. You write a question on the page in ink. A few seconds later, an answer appears — handwritten, in a script that matches the page, flowing under your own words like a reply. The GitHub repo hit 356 points on Hacker News with the on-the-nose framing: it's Tom Riddle's diary from *Harry Potter*, minus the Horcrux.
The stack is unglamorous in the best way. The reMarkable exposes a filesystem over SSH; Riddle watches for new stroke data, ships the page image to an OCR pass, hands the extracted text to a Fable agent, gets a response back, renders that response as strokes in a handwriting font, and pushes it back to the device as a new layer. No custom silicon. No new OS. No cloud-locked SDK you have to pay for. Just a device the author already owned, a Python glue script, and an agent framework the same author happens to maintain.
What makes it feel magical isn't the model — it's that the entire input/output loop stays inside a single physical object you were already going to hold. There's no phone in the loop. No screen tap. No wake word. You write; the page writes back.
The AI-hardware discourse of the last two years has been dominated by devices that failed by trying to be new categories: the Humane Ai Pin, the Rabbit R1, the Friend pendant. Each one bet that a novel form factor would carry a mediocre software loop. Each one shipped a demo that didn't survive contact with a real week of use. Riddle inverts that bet. The hardware is boring on purpose — a $400 note-taking tablet — and every dollar of novelty goes into the interaction.
This is a pattern worth naming. Call it *parasitic AI hardware*: pick a device with a devoted user base and a hackable surface, then graft an ambient agent onto the workflow that device is already good at. reMarkable users are already writing longhand for hours a day. Adding a conversational partner to that surface costs the user zero behavior change. Compare that to the Ai Pin, which asked you to project a laser onto your palm in public.
The technical tradeoffs are also more honest than the pitch decks suggest. Handwriting OCR on cursive is still the weak link — every reMarkable hacker who has tried this hits the same wall around messy input. Latency on e-ink is bounded by the panel refresh, not the model; even a snappy Haiku call feels slow when the display takes 800ms to redraw. And the reMarkable's SSH access is a gentleman's agreement, not a supported API — one firmware update from being closed off. The author is basically betting that reMarkable stays hacker-friendly long enough for this pattern to spread, which is a bet several jailbreak communities have lost before.
The Hacker News thread converged on something interesting: half the comments were about the Harry Potter aesthetic, and half were quietly asking *where's the local model version*. That's the real signal. A diary that phones home to Anthropic every time you jot a thought is a diary your therapist would tell you to burn. The version of Riddle that runs a 4B-parameter model on a Jetson in your desk drawer is the one people will actually use. The current implementation is a proof of concept; the interesting engineering work is the offline variant.
If you build developer tools, the lesson is that the agent loop is now a UX primitive, not a product. A junior dev with a weekend and an agent framework can wire an LLM into any device with a filesystem and a display. That means the moat isn't "we have an agent" — it's the loop tightness: how few milliseconds and how few taps between intent and response. Riddle wins because there are zero taps. Your CLI tool that requires a `--prompt` flag and a 3-second cold start is losing to a piece of paper.
If you're picking a framework for agent orchestration on a side project, this is a decent argument for the small, hackable ones. Fable is a few thousand lines of Python. The author could reason about the whole call graph, which is why he could plug it into a device it was never designed for in a weekend. LangChain, LlamaIndex, and the enterprise stacks give you more batteries — but you pay for it when you want to fit the loop inside a constrained runtime. For hobbyist hardware hacks, pick the framework you can read in an afternoon, not the one with the most stars.
And if you sell hardware, look at reMarkable's incidental win here. They didn't ship this feature. A user did, for free, in public, and every developer on Hacker News now associates the device with a delightful AI interaction. The company's response should be to publish an official plugin API before some overzealous firmware engineer breaks the SSH surface Riddle depends on. Devices that lean into being hackable — Framework laptops, Steam Deck, reMarkable — accumulate this kind of organic demo library. Devices that don't — every locked-down consumer AI gadget of 2024 — die in the review cycle.
The next six months of ambient-AI hardware won't be won by a new device. It'll be won by a dozen more Riddles: an agent grafted onto a Kindle, a Boox, a Playdate, a Freewrite, a Timex Datalink. Each one will be a weekend project. Each one will be more useful than a $700 pendant. The winners will be whoever ships the offline version first, because the diary that keeps its secrets is the only diary anyone actually wants.
Ironically, Tom Riddle's diary encouraging kids to commit suicide [1] and then showing them how to do it would be more on brand than the version in the books.[1] https://en.wikipedia.org/wiki/Deaths_linked_to_chatbots#Suic...
Why so many negative comments on this? I find it cool. I showed this to my son and he absolutely loved it. The author is clearly a fan and this by no means is a production-grade product. Just someone tinkering in their free time and gave the DIY guide away for free. I just don’t understand the negat
I understand that this is kind of beside the point, but it seems like a bad thing to compare inventions to haunted artifacts that mind controlled their users into betraying their friends to a powerfully evil being. (Though since it is being powered by GenAI, which has also driven people to do bad th
This is sick, in the 90's Tony Hawk sense of the word.I love that people can just bang stuff into existence now.There were times in my life where I would wait for an engineering team to change the color of a button for a day to a week.We are not in the slow times anymore.
Top 10 dev stories every morning at 8am UTC. AI-curated. Retro terminal HTML email.
In case anyone else wanted to get to a demo quickly:https://xcancel.com/MaximeRivest/status/2073544461473169432#...Original link for those who want it: https://x.com/MaximeRivest/status/2073544461473169432