Insurgently
Obsidian plugin · Free

Learning Editor

An editor that marks your hedges, filler, typos and repeats and learns what to leave alone.

Learning Editor is the newest plugin in the family. It runs on Model Hub, which holds your AI keys and limits for every plugin. Library Hub turns your books into notes and Topic Map finds the topics across them.

Feedback on your writing, marked on the exact words, from a coach that learns your habits. Nothing changes without your say.

Feedback in the note you're writing

Most writing feedback lives somewhere else. You paste a draft into a chat or a grammar app, read a list of suggestions there, then make each change by hand back in the note. Next time, it flags the same phrase you already said was yours. Here the marks sit on the words in your note, and when you turn one down you can say why, so it stops asking.

A draft about Walden with its marks: words to cut struck through, a doubled word, a long sentence and a vague “this” underlined. The card says why, and the strip at the foot counts the marks by kind.Demo data

What it does for you

A second reader for your drafts. It points at the exact words, says why, and leaves every decision to you.

Marks on the exact words

I wanted feedback I could act on without hunting for it. Each mark sits on the words it means: a wavy underline in its kind's color, or the words to cut struck through in place. Hover one and a card says what to change and why, with a Careful: line when a fix would touch a number, a name or a quote. There are five kinds, Voice, Clarity, Line, Cuts and Mechanics, and you can hide any of them.

Nothing changes until you decide

Apply is one click, and one undo takes it back. Apply all visible takes every safe fix as one change, and skips any that would touch a number or a word you keep. If you've edited the words since a mark was made, it turns gray and won't apply, so a fix never lands in the wrong place.

It learns from what you turn down

This is the part I built it for. When you dismiss a mark, you can say why in one tap: Voice, Wrong or Not now. A dismissed mark stays dismissed, so the same words aren't flagged again.

A profile you write and own

Your profile is an ordinary note: your voice to keep, words never to fix, house rules, habits to work on, suggestions you've turned down. Every check reads it first, so what you keep is never marked. As your reasons pile up, it proposes changes to your profile, like keeping a phrase as your voice or making a habit an automatic check. You accept the ones you agree with. It never edits your profile by itself.

Demo data

Your habits, checked for free

Run Check this note and it marks hedges, filler, close repeats, the words you lean on, very long sentences and sticky ones, heavy with glue words like of, the and that. Harper checks spelling and grammar, bundled with the plugin. It all runs on your computer and costs nothing, and once you've checked a note, it stays up to date as you write. Your own habit rules are small notes, each with examples it has to pass.

A coach for a closer read

When I want judgment, not counting, I ask the coach. Coach this note sends the draft and your profile to a model through Model Hub. It reads the whole piece, marks the places that need work on the same card as everything else, and ends with a short summary and a score out of 10. Each run has a spending limit, $1 unless you change it, and a Stop button.

Under the hood: each run is a Model Hub agent loop with read-only tools (the draft, your profile and a count of your habit words) and a finish tool that hands in the findings. Each finding is checked against the note before it shows, and one whose words can't be found is dropped. Technical details →

Demo data

See whether it's working

The panel shows your habit rate for the open note: habits per 100 words, next to your average across your other notes, with the habits behind it and the hardest paragraphs. It's a count, not a grade, and some of what it counts is your voice. Save the rate to a note's properties and you can follow it in a Base.

Demo data

Everything in one panel

One panel beside the note holds every action: check, coach, clear, hide a kind, apply all, review what it learned. The command palette keeps eight commands, and vim keys work when vim mode is on.

It uses the AI you already set up

Learning Editor holds no keys and picks no models. The coach asks Model Hub, so it runs on whatever you've set up there: a frontier model from Anthropic or OpenAI, a cheap one through OpenRouter, or a local model in Ollama that costs nothing per run. It starts on Model Hub's Careful tier, and you can move it.

Each coach run goes in Model Hub's ledger with every other plugin's calls, so you can see what it cost.

The marks never touch your note. They live in plain files in a .learning-editor folder in your vault, next to your dismissals and the log of your decisions.

Money and data, plainly

  • Learning Editor is free, and staying free. The habit checks, spelling and grammar run on your computer and cost nothing.
  • The coach costs what your provider charges for one run, billed to your account with them through Model Hub. A run stops at its spending limit or after 8 steps. A local model costs nothing per run.
  • When you coach a note, its text, your profile and a count of your habit words go to Model Hub, and on to the provider you chose. Nothing is sent at any other time.
  • Your note changes only when you apply a mark. Your habit rate goes into a note's properties only if you switch that on.
  • Delete the .learning-editor folder and it starts over.
  • I get nothing from you: no telemetry, no account, no ads.
Every request, spelled out: Privacy and costs →

What it can't do yet

  • It's desktop only.
  • A mark has one suggestion, and you can't edit it before you apply it yet.
  • A spelling mark can't add a word to your dictionary yet. The dictionary is a note you edit yourself.
  • It won't rewrite a paragraph for you. It's a coach, not a ghostwriter.

What I'm building next

Learning Editor marks your habits on the exact words and learns what to leave alone. Here is what comes next, with no dates.

Built so far: marks on the exact words, nothing changes until you decide, it learns from what you turn down, a profile you write and own, your habits checked for free, a coach for a closer read, see whether it's working, everything in one panel.

Soon

  • Margin notes. A third way to show marks, beside the text instead of in it.
  • Reader reactions. What a reader takes away from a draft, and where they stop reading.
  • Edit before you apply. Change a suggestion's words, then apply your version.
  • Your words, kept visible. See which words came from a suggestion, so a draft stays plainly yours.
  • Settings for one note. Switch it off for a note, or give a note its audience, its purpose and a sample of the voice you want.
  • The cost before the coach runs. See what a run will cost before it starts.

Later

  • Practice drills built from the marks you applied and dismissed.
  • Fact checks for the claims the coach flags, using web search through Model Hub.
  • Shared rule packs. Import habit rules other writers share.
  • Retire a habit you've beaten: it stops marking it and keeps counting it.
  • Comments as work items. Leave a comment in a draft, and an agent picks it up as a task.

Get it running

You need desktop Obsidian 1.13.1 or later, and Model Hub with one provider set up. Both are in Obsidian's community directory.

  1. Open Settings → Community plugins and select Browse. Search for Learning Editor, then select Install and Enable. No Model Hub yet? Install it the same way and add a provider.
  2. Select the spell-check icon in the left ribbon to open the panel, then select Create your profile.
  3. Open a draft and select Check this note. For a closer read, select Coach.
Getting started, step by step →

Curious how it works inside? Start with Technical details →. Want an editor of your own on the same code? The Developer docs → show how.

Join

One list for everything here: an email when a plugin ships, plus the posts I write about these projects on brianspringer.com.

Want to help build this, or join the co‑op? Email me about joining →

Insurgent Methods is where we build at a different scale, to take on the biggest problems.

Needs
Obsidian 1.13.1, desktop