Omahub
← All plugins
L

Omoide

by leweyse

Capture a screenshot, note, or voice memo and let an agent turn it into a structured memory with reminders, events, and collections.

GitHub

Security review

Review recommended · 2 findings

Deterministic scan — not a security guarantee

Low
Risk level
Low
Analyzed commit
265683d
Scanned
4 days ago
  • low obfuscation dev/parity:128

    Augments a command with octal/hex escape sequences.

    \x89PNG\r\n\x1a\n":
  • low obfuscation dev/parity:216

    Augments a command with octal/hex escape sequences.

    \x89PNG\r\n\x1a\n"

Automated analysis only — not a security guarantee.

AI advisory review

No obvious issues detected

Language-model assessment · ~deepseek/deepseek-v4-flash-latest — advisory only

Low
AI risk level
Low
Recommendation
install
Model
~deepseek/deepseek-v4-flash-latest
Analyzed commit
265683d
Reviewed
4 days ago

The deterministic scan's two 'obfuscation' findings are false positives: dev/parity contains PNG magic bytes (\x89PNG\r\n\x1a\n), which are the standard file signature for PNG images in a test fixture, not obfuscated code. The sampled QML code is clean and follows good practices — subprocess calls use argv arrays rather than shell strings, and the CLI is built from source on first run rather than shipping a precompiled binary. The plugin's behavior (sending captures to a user-configured AI agent, running custom agent commands via sh -c) is documented and user-controlled.

  • The cli/ source directory was not included in the review sample; the deterministic scan found nothing malicious there, but a human may want to verify the C code that gets compiled and executed on the user's machine.
  • Custom agent commands are executed via sh -c, so a malicious or compromised agent configuration could run arbitrary commands — this is documented and user-configured, but worth being aware of.
  • The plugin sends screenshots and notes to a user-configured AI agent endpoint, which is a privacy consideration inherent to the plugin's stated purpose.
How this check works

This review combines the deterministic scan (the rule-based results above) with an independent look at the plugin's code by a language model. The model reads a trimmed sample of the repository's files, the manifest, and the README, then gives a plain-language risk level and a recommendation: install (no notable danger), review (look closer first), or avoid (clearly dangerous).

It runs on the same analyzed commit as the deterministic scan and is strictly advisory — it is not a security guarantee and never blocks a plugin by itself. A human moderator still reviews plugins before they are listed.

AI advisory only — automated analysis, not a security guarantee.

Install
$ omarchy plugin add https://github.com/leweyse/omoide --enable
Productivity #ai

Omoide

思い出 (omoide) is Japanese for a memory you keep.

The For you view, listing today's events and the open tasks

More in assets/: the library, the capture menu, agent settings.

The idea is Nothing's, from Essential Space on their phones. I wanted it on my computer, which is where I actually work, and Omarchy's plugin system is what made that possible. Grab a screenshot, type a note, or dictate one. An agent turns it into a memory with a title, a summary, key facts, dates and to-dos.

Everything runs on your machine. No account, no server, nothing syncing. Your memories are a SQLite file and a folder of images in your home directory. The only thing that ever leaves is what you hand to a cloud agent, and nothing at all if you run a local one.

Demo

https://github.com/user-attachments/assets/f6019266-45bc-47b5-b1b4-4b8119e7e821

Capturing

  • Left-click the bar icon opens the menu.
  • Right-click opens the library.
screenshot ──▶ pick a region ──▶ overlay opens with the shot, note optional
note ────────▶ overlay opens empty, text required
voice ───────▶ overlay opens already listening, transcript lands editable

Enter saves. Esc discards the draft and its image. The button at the end of the note field starts dictation in any mode, if Voxtype is installed.

Browsing

One dialog, three views.

For you is today: a count of events and open tasks, then both lists.

Library is every capture, newest first, in a masonry grid. Agent tags filter it. Search matches titles, notes, list items and OCR'd text, by prefix.

Tasks has three tabs, Upcoming, Past and Completed. Space ticks a row or accepts a suggestion.

Keyboard

The dialog works without a mouse. Ctrl 1 to 3 switch views, / searches, Space ticks a task or accepts a suggestion, Enter opens what the cursor is on, Esc steps back and then closes. ? shows the full map.

What a memory holds

A memory is an ordered list of typed blocks, and which blocks appear depends on what you captured. An article gives you a source, a summary and key points. A screenshot of a design board gives you an image and a list of names, nothing more.

Headings, list items and to-dos are editable, so you can fix what the agent got wrong. Related captures and collections are yours to link, never inferred.

Tasks, events and reminders

Open any task or event to edit it: title, date, time, and as many reminders on one item as you want. Nothing commits until Save, so closing the sheet drops a mistyped date.

Reminders are armed inside the shell from the database, so they survive a reboot and need nothing outside Omoide. One that came due while the machine was off still arrives, if it is less than half an hour late.

Anything the agent inferred rather than being told arrives as a suggestion, a bullet with a + and no alarm until you accept it.

Install

omarchy plugin add https://github.com/leweyse/omoide.git --enable

It asks which bar section to put the icon in. The CLI is C, and there is no binary in the repository: the shell builds it from cli/ the first time it loads the plugin, with the clang and libraries Omarchy already has, in a second or two. An update rebuilds it the same way. Screenshots need grim and slurp, OCR needs tesseract, dictation needs voxtype (omarchy-voxtype-install).

To update, omarchy plugin update leweyse.omoide, then omarchy restart shell. The shell rescans plugins on update but keeps rendering from its cached QML until it restarts, so without one you are still running the old version.

To remove it, omarchy plugin remove leweyse.omoide. Your memories stay on disk. Run ~/.cache/omoide/bin/omoide uninstall --purge first if you want them gone with it.

For keybindings, in ~/.config/hypr/bindings.lua:

o.bind("SUPER + CTRL + <YOUR_CHOICE>", "Omoide capture",
       "omarchy-shell -q omoide toggleChooser")
o.bind("SUPER + CTRL + <YOUR_CHOICE>", "Omoide library",
       "omarchy-shell -q omoide toggleSpace")

The capture binding opens the capture options in Omarchy's own menu, default action first and focused, so Enter starts it. No binding starts a capture directly.

Choosing an agent

Presets cover claude, codex, opencode, gemini, ollama and aichat. custom takes any command that reads the prompt on stdin and prints the answer on stdout.

Screenshots reach the agent as OCR text by default, so the image itself stays here. "What the agent sees from a screenshot", in agent settings, switches that to the screenshot itself, which reads charts and dense pages far better. Only claude, codex, gemini and opencode take images, and the card says so for the rest.

Presets run with their tools switched off, in an empty working directory: a capture is a transcription of whatever was on screen, which can include a paragraph written to be read by whatever handles it next, and these are coding agents with a shell. Omoide passes each CLI's own flag for this, so claude gets no tools, codex its shell tool switched off inside a read-only sandbox, gemini its read-only mode, and opencode a deny-all permission set merged over your config. custom is run exactly as you wrote it, by sh -c, so quotes and variables mean what they do in a terminal; add your agent's own read-only flag to the command yourself. The prompt reaches it on stdin, never on the command line.

With no agent you still get the capture, your note, search over OCR'd text, and a reminder from an explicit "remind me to call the vet tomorrow at 3pm".

Where things live

~/.config/omoide/config.json     the agent you chose
~/.local/share/omoide/           memories.db and blobs/, your captures
~/.local/state/omoide/           derived cache, rebuildable
~/.cache/omoide/bin/             the CLI, built from this checkout

The plugin writes nothing else, unless you run omoide install, which also links ~/.local/bin/omoide to the CLI so you can run it by name. XDG_*_HOME moves the base directory, but the omoide/ leaf is always hardcoded, so uninstall can only delete a directory the plugin made itself. No environment variable hands it a path to remove.

TODO list

Voice dictation. Voxtype does the transcription and the wiring is in, but I have not put real use behind it. Next thing I want to fix.

Export memories. I still capture things on my Nothing phone and I want the two halves to meet without a cloud account in between. The phone end is the part I have not solved.

Calendar integration. Several providers, ideally, though it is low on my list.

Clipboard capture. It already shows in the menu, disabled.

IPC and trust

IPC opens windows, it does not act. A capture starts only from a click or Enter on a shell surface. Nothing runs OCR or reaches an agent until you press Save, IPC cannot start the microphone, and Esc deletes an unexpected capture, image included.

License

MIT, in LICENSE.