Remote Frog

The need came up while I was helping the folks at Corolario AI, more specifically Marcos, prepare a proposal for a construction company.

On a construction site, things happen all the time: decisions, incidents, small discoveries that matter, but most of them rarely end up captured in ways that are manageable and reusable. The challenge was how to let site managers capture what they see, decide, and discover directly into a company’s centralized brain so that others can directly use this knowledge.

The non-functional requirement was clear: it had to be lightweight. No extra app, no extra form, no extra layer of process that someone would have to review later. Because every extra step is also a place where knowledge can end up waiting around. And die of old age.

The solution came to me while I was thinking about another Obsidian-based app I’d been refining: a working-memory tool for people with ADHD that I called Jiminy Cricket.

In practice, I only needed to package one of its pieces to capture text in a simple but robust way, especially on mobile. In other words: make it possible for a site manager to dictate what they see into their phone while they’re on a visit.

The idea: speak instead of writing

If the company already uses Obsidian as its knowledge repository, why not let someone simply start talking about what they’re seeing and have that generate a note with the right metadata in the shared institutional memory?

This is Dictation, a speech-to-text plugin for Obsidian that works on both desktop and mobile. The flow is deliberately simple:

• Open the note where you want to capture something.
• Press the microphone and speak.
• Press stop when you’re done.
• A floating menu lets you choose: keep the audio, transcribe it, or delete it.
• If you transcribe it, the text appears in place, and the audio can be kept or discarded depending on your settings.

Better to see it.

What else the plugin includes

• Near-zero friction. What matters here is not just transcription, but the flow design: capture an observation in context, with as little friction as possible, and return it exactly to the place where it should live. Dictation stores the note and cursor position when recording starts. Even if you move around Obsidian or switch notes, it will try to return the transcription to the original destination.
• Choice of provider and model. The plugin lets you choose between OpenAI, Groq, Deepgram, or any OpenAI-compatible endpoint. It also lets you pick the language, auto-detect it, and configure the model through a small setup assistant.
• A mobile experience that’s been especially cared for. On mobile, it shows a floating recording button that stays accessible when the keyboard appears. On desktop, you can start it from the sidebar, the status bar, the command palette, a keyboard shortcut, or push-to-talk.
• Focus on robustness. I also wanted to take care of failure cases and make the plugin more robust, especially in an area where I found similar options in the Obsidian library tended to fall short:
  • If transcription works but the destination can’t be found, it shows the text so it can still be copied and recovered.
  • After recording, you can transcribe, keep, or discard the audio.
  • After a successful transcription, the audio can be moved to a managed folder and deleted after the configured period. By default, 21 days.
  • The API key is stored in the device’s secure storage.
  • There’s no telemetry, and note contents are not sent to the provider. Only the audio and the options needed for transcription are sent.

It’s also worth being clear about what it is not: Dictation uses whichever transcription provider the user chooses, so it is not a fully local or offline solution. That said, if you’re offline, it can still save the audio and let you transcribe it later.

Weren’t there already plugins for this?

Yes. There are several, and some of them do quite a bit more.

Local Dictation, for example, focuses on local models and adds features for meetings, system audio, speaker identification, and timestamps. Whisper can record or transcribe existing files. Other plugins offer voice commands, post-processing with language models, or much more complete recording workflows.

Dictation is not trying to win that feature race.

Its bet is more focused: capture an idea in the most direct way possible, whether from desktop or mobile, place it safely where the recording started, and give the user explicit control over the provider and the audio destination.

The hypothesis behind the plugin: SAME FLOW, REDUCE FRICTION

The hypothesis is simple: for many adoption problems, you don’t need a new platform; you need to reduce friction so knowledge can be captured inside the system the organization already uses.

For a site manager, that could mean tapping a button on their phone, dictating an incident, a decision, or an observation, and then getting on with their work. No WhatsApp, no recorder app, no form, no extra app in the middle.

That’s what I wanted to test: whether a small solution, integrated into the real flow of work, can be more useful than yet another tool the team has to learn, open, and maintain.

The best part: from AI Enablement to a real prototype

The way of responding to this problem comes in good part from an Automattic experience I took part in and found incredibly powerful: AI Enablement.

We spent two weeks in New York, around fifty people from product, design, engineering, data, and operations, combining talks, workshops, and, above all, time to build prototypes. There’s a good recap of those two weeks on Data.blog. Short version: it was intense, thoughtful, and just plain great. The kind of experience that reminds you AI is not only about consuming tools, but about learning how to think with them, prototype quickly, and turn vague ideas into things that actually work.

One of the most valuable things was sharing space with people who had different problems and perspectives, showing something still imperfect, and then continuing to build from there. AI makes it much easier to reduce the distance between spotting a problem and having something that works well enough to test. And this plugin, precisely, comes from that.

An open experiment

Dictation is still an exploration. It’s not a deployed solution for that construction company, and it doesn’t pretend to solve knowledge management for a team on its own. But it does test a hypothesis I care about: maybe many adoption initiatives do not need another platform. Maybe they need a more natural entry point into the system the company already has.

If this idea works, what will matter is not only that I built an Obsidian plugin that others can reuse, but that it helps prove something more general: many times the problem is not a lack of knowledge, but the distance between living it and capturing it.

The plugin is free and open source. You can install it from the Obsidian community plugins, and the code is available on GitHub. If you work with construction, maintenance, operations, or any other environment where note-taking competes with having your hands busy, I’d love to hear whether this flow fits reality. Or where it breaks.