Obsidian Stats logoObsidian Stats

AI Assist

by mrrepac
5
4
3
2
1
Score: 50/100
Install Code

Description

Obsidian plugin: sidebar AI chat and in-place rewriting of the selected text

Reviews

No reviews yet.

Stats

0
stars
52
downloads
0
forks
15
days
0
days
0
days
0
total PRs
0
open PRs
0
closed PRs
0
merged PRs
0
total issues
0
open issues
0
closed issues
45
commits

Latest Version

0.10.0
19 hours ago

Changelog

A version number for the data files, a panel that starts clean, and settings that survive the trip between machines.

What changed in how it behaves

Two defaults are different from 0.9.0. Both can be turned off in the settings.

A run from a note starts a new conversation. Every path counts: the quick menu, a one-off prompt, "ask about the selection", and — new in this release — an action that rewrites the text in place. The panel is cleared first, so the card you are waiting for is the only thing on screen, and the conversation before it is not paid for again on the next question. The cleared conversation comes back if you click the notice, which stands for eight seconds.

The edit log is cleared along with it. The cards used to survive because each one carries an "Undo edit" button. But the text is rewritten in the note itself, and Ctrl+Z was never taken away. If you want the button within reach anyway, turn on Keep the edit log — it sits under the toggle above and appears only when there is a clearing to be kept from.

The data files have a version now

data.json and history.json carry a schemaVersion, and the format they describe is written down in docs/data-format.md. Nothing about this is visible in use; it matters because after 1.0 these files may not be broken, and until now there was no number to check.

Four outcomes are told apart when a file is read: absent, current, written by a newer version of the plugin, and unreadable. The interesting one is the last.

A settings file that will not parse is copied aside, not silently replaced. Obsidian's own loadData() returns nothing both when the file is missing and when it is corrupt, and the plugin used to treat the two the same — which means a stray byte in data.json cost you every API key you had entered, without a word. The broken file is now copied next to the original under a free name with a timestamp, and you are told where it went. The chat log gets the same treatment, quietly: there are no keys in it.

Settings arriving from the other machine are no longer trampled

If your vault syncs, data.json written on another machine used to be overwritten by whatever this machine happened to hold — and since a draft in the input field is saved as you type, the overwriting was armed on every keystroke. The plugin now notices the file changing underneath it and re-reads the settings.

Only the settings: re-reading everything would throw away a live conversation. And only when the file parses — a file that arrives broken, or vanishes mid-read, leaves what you have alone.

Worth knowing: Obsidian Sync carries manifest.json, main.js, styles.css and data.json, and nothing else. history.json — your conversations — stays on the machine that wrote it.

Under the hood

455 checks, up from 412. Reading and writing the two files moved into store.ts, which imports nothing from Obsidian, so the tests can drive it directly against deliberately damaged files. The rules for what survives a clearing and what a restore keeps live in history.ts for the same reason.

One race is fixed along the way: the undo notice lives for eight seconds — exactly as long as the model spends rewriting — and clicking it used to swap the whole feed for the snapshot, carrying off the card of the edit in flight along with its undo button, while the note had already been rewritten. The restore now merges the snapshot with whatever arrived after it.

See all version on GitHub

README file from

Github

RU AI Assist

A sidebar chat with a language model, and in-place rewriting of the text you selected.

Built for people working from Russia. Every provider it ships with is reachable without a VPN and takes Russian payment cards: DeepSeek directly, plus three Russian aggregators — Polza.ai, ChadGPT and GPTunnel — which give you GPT, Claude, Gemini and the rest through a single key. And Ollama or LM Studio, if you would rather run a model on your own machine.

Most AI plugins are built around the chat. This one is built around the note: you select a piece of text, press a key, and the text changes. The chat is there too, in the right sidebar, but the main road is selection → key → done.

Читаете по-русски? Есть русская версия README.

What it does

  • Actions over the selection. Fix spelling, improve the text, expand it, shorten it, evaluate it — or write your own prompt. The result replaces the selected text; the undo is the usual Ctrl+Z.
  • Selecting is optional. The settings decide what an action works on when nothing is selected: the whole note, the section under the cursor (from its heading down to the next one of the same level, subsections included), the paragraph — or nothing at all, if you would rather always be asked to select.
  • Quick menu opens over the selection: the digits run your presets — five keys to begin with, up to nine if you add them — and the field below takes a one-off instruction. The digit only picks a preset while the field is empty, so a prompt can start with a number. The answer to a one-off instruction comes to the chat; Ctrl+Enter puts it into the note instead. The physical key is what counts, so any keyboard layout works. The menu comes without a key of its own: switch on Alt+1 in the settings, or bind whatever suits you.
  • Nothing selected — you see what was taken. The menu then grows a row — paragraph, section, note — with the size of each, the one from your settings picked and the piece itself highlighted in the note. The arrow keys change it for this one run; the setting stays as it was.
  • Right-click on the selection. The editor's context menu offers the quick menu and a question to the chat — and the actions themselves, each on its own line, if you want them there. On a phone it is the same menu, held down. With nothing selected the plugin adds nothing to it.
  • Ask about a fragment. Select a piece of a note and go to the chat: it comes along as an attachment, so the question can be just "and shorter?" — and the fragment stays with the conversation for the questions that follow.
  • An image with the question. A screenshot from the clipboard (Ctrl+V), a file dragged straight into the panel, the paperclip in the footer — or a right-click on it: "from the vault" and "from this note", by the images embedded in it. Before it is sent the image is scaled down to a sane size, its weight is shown on the chip, and it travels along with the question. Images from earlier replies are not carried into a new request: each one is paid for on every question. A model that does not take images says so before you send, rather than by refusing afterwards.
  • A PDF in the question. A document is attached the same ways a picture is, but travels as text: the plugin reads the text layer — the one already inside the PDF — and sends it along with the question. So it works with any model, including the ones that cannot see pictures. The chip shows up front how many pages and characters are going; a long document is sent from the beginning, and it says so right there. A scan has no text in it at all, and the plugin says plainly that there is nothing to pull out. It brings no reading library of its own: it uses the one Obsidian's own PDF viewer runs on.
  • An action can have its own model. Proofreading is done just as well by the cheapest one, and judging a text without a strong one is pointless — pick a provider inside the action and it always goes there, whatever stands in the panel header.
  • A cut-off answer can be finished. The model ran out of room and stopped mid-word — "Finish it" asks for the rest and joins it to what you already have. Works both in the chat and for an edit over the selection: the half you paid for is not thrown away.
  • A journal of edits in the side panel: what ran, what came back, what exactly changed (word-level diff, "was → became"), an "Undo the edit" button and a "Run it again" one — it brings the text back and runs the same action over it once more.
  • Chat in the right sidebar with streaming, thinking blocks, the running cost in tokens, and a button that saves the conversation into a note.
  • Sources under the answer. A search model answers from pages on the web rather than from memory — the list of them sits under the answer, collapsed, and opens on a click. It is stored with the answer, so it survives a restart, and goes into the note if you save the conversation. The [1][6][8] tails such a model puts after every sentence are stripped from the text — they get in the way of reading, and the list is right there.
  • A private chat can have its own model. It is started to ask something that is not about the vault, and the model wanted there is usually a different one — pick it in the settings and the panel's own stays where it is.
  • The conversation can be rewritten. Under an answer — "ask again"; under your own question — a pencil that puts the text back into the field, so the question can leave as a different one. The bin removes what is not needed, and whatever was removed comes back from the notice.
  • The input field remembers. An unfinished question survives closing the panel and restarting Obsidian, and the up arrow walks through your previous questions.
  • The model can edit the note itself through tools — every edit is shown as a card with an Apply button, so nothing lands in your text without you. An action that answers in the panel gets no tools at all: "show it in the panel" already says "leave the note alone".

Providers

Anything that speaks the OpenAI format. The presets are the ones that work from Russia without a VPN:

  • DeepSeek — the Chinese model directly: cheap, no middleman;
  • Polza.ai, ChadGPT, GPTunnel — Russian aggregators: one key, and behind it GPT, Claude, Gemini and dozens of other models, paid for with a Russian card;
  • Perplexity — less a model than a search engine with a model on top: it reads the web first, then answers and attaches the links. You reach for it for facts and fresh news — the things an ordinary model does not know and happily invents. It has no note-editing tools, and the plugin does not send it any;
  • Ollama and LM Studio — a model on your own computer, free and offline. With LM Studio the answer arrives whole rather than letter by letter: it does not let a plugin read it as it is written.

Plus any other base URL you type in. The API key, the model and the address are remembered separately for each provider, so switching back and forth does not lose your keys.

With Ollama or LM Studio running on your machine, nothing leaves it at all: no key is asked for and none is sent.

Getting started

  1. Install the plugin and open its settings.
  2. Pick a provider, paste the API key, and choose a model — the "Fetch" button asks the provider for the list. There is a "Test connection" button right below.
  3. In the "Quick menu" section, switch on Alt+1 — or give the "Quick menu over the selection" command a key of your own in Obsidian's hotkey settings.
  4. Select some text in a note and press it. Or right-click the selection — the plugin is already there.

Settings

Everything lives in three sections:

  • Provider — service, address, key, model, connection test.
  • Behaviour — temperature, response length limit, streaming, what to do when nothing is selected (the whole note / the section / the paragraph / nothing), the length past which it asks before sending, token usage display, the model for a private chat, where new notes go (the vault root / a folder of your own / next to the note being worked on), whether images brought into the chat are kept in the vault, note editing tools, the system prompt for the chat. The same section decides how the panel starts: whether it opens empty when Obsidian starts, and whether it is cleared before every run from a note (and if it is — whether the cards of edits already made, with their “Undo edit” button, survive the clearing).
  • Quick menu — the keys: five by default, more if you add them, and a row can be dragged by its handle; plus what the plugin puts into the editor's context menu. Each key is an action, and this is where an action is edited: name, icon, prompt, what to do with the answer and which model runs it. Your own action is created in a slot too. An action taken off a slot is not lost — it keeps its own command, and a command can be given a hotkey.

Commands

Every action is a command, so any of them can get a hotkey of your own. Plus: open the chat, start a private chat, ask about the selection, repeat the last action, stop generation, and the quick menu. The last two take Alt+1 and Alt+2 if you switch that on in the settings.

What is sent where

The text you run an action on goes to the provider you configured, and nowhere else. The plugin has no telemetry, no analytics and no server of its own; the settings live in .obsidian/plugins/ai-assist/data.json and the chat history in history.json next to it, inside your vault.

An image attached to a question goes to that same provider and inside the request itself, as a data: address, without being uploaded anywhere. With "Keep images brought into the chat" switched on it lands in the vault's attachment folder as the question is sent — the same folder Obsidian itself uses; switched off it lives until the next restart and never touches the disk. In a private chat it is never saved.

From a PDF it is the text that goes, not the file: the plugin reads the text layer and sends it to the same provider the question goes to. The document itself is not copied into the vault — the setting about images does not cover it.

Two things send more than the question itself. "Send the current note as context" attaches the open note to every request in the chat. And a model allowed to edit notes can read the open one whenever it decides to — that is a separate switch, and every call it makes is shown in the panel as a card.

When none of that is wanted, there is the private chat — the "New private chat" entry in the header's … menu, the command in the palette, or Alt+2 if you let the plugin take the keys. It works like a private window in a browser: a separate conversation where nothing leaves the vault at all — no note, no selected fragment, no tools, no system prompt. Just your question and the answer, the way it works on the provider's own site. The header is tinted there, and in place of the context button sits an unplugged cord: there is nothing to switch on. A normal "New chat" brings everything back.

Requirements

Obsidian 1.8.7 or newer. Desktop and mobile. An API key from one of the providers, or a local server on your own machine.

License

MIT — see LICENSE.

Sponsors