QnALog

English | 简体中文

Turns conversations into structured knowledge.

Open-source conversation intelligence for Obsidian: record, transcribe, and organize meetings, interviews, talks, and voice notes into Markdown you can reuse.

QnALog ships no API keys and works without an account: you configure your own speech-to-text (ASR) service and, optionally, your own large language model (LLM). Recordings and generated notes stay in your vault.

Supports desktop and mobile Obsidian. Mobile recording uses the device microphone. System audio, virtual audio devices, multichannel capture, desktop device diagnostics, and realtime streaming ASR providers that need custom authentication headers require the desktop app.

Relationship to LexVoice

Origin. QnALog is derived from LexVoice by Lynnx, based on its last MIT-licensed release (2.1.2). Upstream relicensed to a proprietary license from 2.2.0 onward; QnALog is an independent project, not a newer version of LexVoice and not a continuation of that release line. It has since been reworked into a plugin with its own name, data namespace, and settings. See NOTICE and MAINTAINING.md.

There is no data path between the two.

• No migration. QnALog has no import, export, or migration path for LexVoice data. It does not read LexVoice notes, markers, tags, folders, or settings, and it does not scan or rewrite your existing files on load.
• No settings inheritance. A new install starts from QnALog's own defaults. If you have used LexVoice, its API keys, service configuration, folders, and prompts are not carried over — configure QnALog from scratch.
• Its own namespace. Tags, markers, and folders written to your vault use the qnalog / QnALog namespace only.
• Its own plugin id. The id is qnalog, different from LexVoice's lexvoice, so Obsidian manages them as two separate plugins.

Features

Live outline

Chapters grow as you record. After recording, chapters link to the player — click a chapter to jump to that position in the audio. When recording stops, AI completes the chapters into a full set of notes.

From a completed note's sidebar, choose Rebuild outline from all transcripts to regenerate only the outline from the full original transcript ledger, even when automatic live outlines are off. Before replacing the current outline details, QnALog backs up the exact note under <vault>/<configDir>/qnalog-outline-backups/<timestamp>/<filename>. If generation is incomplete or fails, or if the note changes while the outline is being generated, the note is left unchanged. This action does not reorganize the note body.

In-meeting notes

While recording, jot live notes under the outline. The first character can trigger different handling:

Trigger the AI assistant:

• #term — the AI explains the term in the context of the current discussion.
• ?question — the AI answers using the current transcript and outline.
• !highlight — marks something important and has the final notes treat it accordingly.

Mark only (no AI call):

• @assignee — record "@alice follows up"; the final notes prefer assigning that todo to them.
• /todo — capture an explicit todo candidate.

Half-width and full-width symbols are both accepted. In-meeting notes are fed into the final summarization prompt as clearly-labeled "live supplementary material", never mixed into the raw transcript.

Continue a recording into an existing note

Choose Append recording to this note from the open note's sidebar or file menu. The floating bubble checks the note that is active when you click and again after reading it; a delayed read cannot redirect the recording to an earlier note. QnALog records the new audio and transcript in a separate pending note, then merges them into the selected target after its active processing and transcription tasks finish. A complete transcript ledger is sufficient even when the target has no outer transcript-container marker. Failed merges keep the staged note and audio for retry; the progress window shows the active step, failure, and retry state. If the target note is missing or its transcript identity changes, the separate recording is kept instead of being written to a different note.

Ask this note

Ask follow-up questions when the final notes miss a detail or you want to revisit a specific part of the discussion. QnALog answers from both the organized note and the preserved raw transcript. Useful answers can be written back to one compact Ask this note section in the Markdown file.

Long meetings & recovery

In standard meeting and learning-note modes, long recordings are organized in recoverable parts instead of relying on one all-or-nothing LLM response. QnALog builds a global topic map, saves each completed part as a local checkpoint, and assembles the final note in time order.

If a request is interrupted or a model reaches its output limit, completed work is reused and only unfinished parts are retried. The raw transcript remains available, and an incomplete result is shown as partially completed rather than being saved as an empty note.

Source-linked notes

Transcribed notes retain the original ASR text and revision history alongside the visible transcript. Decisions, actions, questions, and topics can include references to exact transcript passages. If a passage changes, the saved references are marked stale instead of being presented as current. The evidence is returned with existing organization requests; it does not trigger a separate extraction request. Generate Clean Copy writes a separate derived file and activates that version in the source note without replacing the raw transcript. Regular reorganization creates a derived Markdown file and a minutes cache but does not activate it automatically. Before the first derived version, QnALog stores the safely identifiable visible source text as a source-original cache; if that snapshot or its index cannot be verified, generation or switching stops before changing the source body. Recent Notes shows the original row only when a valid snapshot exists, followed by derived versions; clicking a row displays its content in the source note. QnALog's .versions cache is separate from Obsidian's built-in file Version History: the plugin cache stores selectable versions, while file history records physical Markdown edits and can restore an earlier source body. If a clean copy already exists, Generate Clean Copy offers to show it or regenerate the same file.

Task progress

The processing panel separates transcription, AI organization, and Markdown writing. It shows the active stage, recent activity, failures, and retry or cancel actions. Failed transcription and failed AI organization remain distinct so you can resume from the step that actually failed.

Sediment & object library

After each note, AI splits the content into three candidate groups:

• People — adjudicated one by one: keep, merge, or ignore
• Todos — selected by default
• Hotwords — names, organizations, brands, terms, to improve later ASR accuracy

Confirmed items become standalone Obsidian objects — people profiles, todo cards, and ASR hotwords — and the sidebar can assemble confirmed todos into a todo wall. When the same person comes up again, a note links to the existing profile instead of creating a duplicate.

Todo enhancements

Edit owner, due date and sub-tasks inline at the candidate stage — no dialogs. Stored todos use standard Markdown task syntax (recognized by plugins like Tasks). Source information is preserved on delete / redo for traceability.

Recording reliability

• Level meters before and after recording show whether the mic and system audio are actually working.
• Audio inputs remain user-selectable; virtual or remote device names are shown as guidance rather than being selected or rejected automatically.
• A device check in settings diagnoses "recorded but silent" problems.
• Compatible independent multichannel input can be detected and transcribed by channel, with speaker labels that can be mapped to names. Separation stays off when independent channels cannot be verified.
• Deleting a transcript offers to delete its audio file too.
• Recordings under 10 seconds are not transcribed or turned into minutes. Audio under 3 seconds is discarded; audio between 3 and 10 seconds stays in the recording folder so you can import it manually if it turns out to matter.

Export

From one set of notes you can generate an HTML report, a PDF report, or an .eml email draft — same content, different presentation.

Note list

The sidebar can organize recent notes by folder or by time. Folder groups can be collapsed, the open note is highlighted, and search and template filters remain available in either view.

Basic usage

1. Click the Q-shaped QnALog icon in the left ribbon to open the sidebar. Hover over it to see the QnALog live minutes panel name.
2. Choose a template and an audio input.
3. Start recording; check that the level meter reacts.
4. Watch the live outline; add in-meeting notes if needed.
5. Stop recording and follow transcription and AI organization in Task progress.
6. If processing was interrupted, retry only the failed stage; if the notes miss a detail, use Ask this note.
7. Open Sediment and review people, todos, and hotwords.
8. If you need to share, use Export.

Default folders (all configurable in settings). Names are chosen once at first install, from the interface language:

A Chinese interface gets the left column; every other language gets the right one (only Chinese and English names exist). Changing the language later does not move folders that already exist. Default folders use the QnALog/ prefix; they are ordinary paths and can be changed in settings at any time.

Requirements

Required:

• Obsidian 1.11.4 or later
• A speech-to-text service (cloud API or local)
• A vault folder for recordings and notes

Recommended:

• An LLM service — for the live outline, note organization, sediment, export, and template tuning
• A virtual audio device and a real microphone — to record system / online-meeting audio and your own voice; see below
• A domain glossary — improves recognition of names, products, organizations, and terms

Audio input & real microphone

Capturing system audio cross-platform from the Obsidian desktop app is unreliable, so recording online meetings, web video, courses, or anything played by the computer usually needs a virtual audio device:

• Windows: VB-Cable
• macOS: BlackHole
• Linux: PulseAudio / PipeWire monitor source

On Windows with VB-Cable, mind the naming:

• Meeting apps, browsers, and system output → CABLE Input
• QnALog reads CABLE Output (a recording device)
• To also record yourself, the real microphone must be your physical mic — not CABLE Output, BlackHole, VoiceMeeter, or Stereo Mix

If the level meter does not move, run the device check before starting a long recording.

Privacy, network, and file access

No ads, no analytics, no telemetry. Plugin settings are stored locally in .obsidian/plugins/qnalog/data.json; API keys are stored through Obsidian SecretStorage, not in that file. When using another Obsidian installation, confirm that SecretStorage has the keys and re-enter any that are missing.

Network use. QnALog works offline unless you configure a service that needs the network. When you do, requests go only to the endpoints you configure:

• Speech-to-text requests send audio to the transcription service you configured.
• AI organization requests send transcript text and prompt context to the LLM service you configured.
• The update check requests manifest.json from this project's GitHub release page only when you click Check for updates in Settings > About or run Check for Updates from the command palette. Plugin startup makes no update-check requests. It never downloads or installs anything.

Where recordings live. The current version has no QnALog backend: it keeps no recordings, transcripts, or notes on servers of its own, and everything it writes — recordings and generated notes — stays in the local vault path you choose.

Optional subscription (planned). A paid subscription may be added later for users who would rather not manage API keys: sign in, and transcription and AI organization run through that service without any provider setup. It stays optional. If you turn it on, the audio and text of a recording go to that service instead of your own endpoints; recordings and notes are still written to your vault. If you leave it off, everything described above is unchanged.

Files outside your vault. The optional external inbox feature can watch a folder outside your vault (an absolute path, for example a synced recordings directory) and import audio from it. That access happens only if you configure such a path, and it is limited to reading the files you point it at.

For sensitive content (client data, medical, legal, HR, recruiting, internal strategy), prefer local transcription with a local model, and obtain consent before recording. Details: PRIVACY.md.

Installation

Option 1 — community plugin directory (desktop and mobile)

1. In Obsidian, open Settings → Community plugins and browse the directory.
2. Search for QnALog, install it, then enable it.

The directory's LexVoice entry is a different plugin — see Relationship to LexVoice.

Option 2 — BRAT (works on desktop and mobile)

1. Install and enable BRAT from the community plugin directory.
2. In BRAT, choose Add beta plugin and enter qnalog/qnalog.
3. Install, then enable QnALog under Settings → Community plugins.

You can also download the release assets (main.js, manifest.json, styles.css, plus LICENSE and NOTICE) straight from the releases page into <vault>/.obsidian/plugins/qnalog/. Each release is built from its own tag by CI, which rebuilds the bundle and byte-compares it before uploading.

Option 3 — build from source (desktop)

git clone https://github.com/qnalog/qnalog.git
cd qnalog
npm ci
npm run build
npm run install:vault -- "/path/to/your/vault"

install:vault copies main.js, manifest.json, styles.css, LICENSE, and NOTICE into <vault>/.obsidian/plugins/qnalog/, and snapshots anything it overwrites (including data.json) into <vault>/.obsidian/qnalog-install-backups/<timestamp>/. It only touches the qnalog folder: settings are never inherited from other plugins. The settings folder follows the plugin id, so a build installed from source and a BRAT install keep their settings in the same folder.

Then reload Obsidian (Ctrl/Cmd + R) and enable QnALog.

QnALog keeps your settings when the stored schema is older: it migrates them forward and your API keys, service configuration, prompts, paths, and device choices are preserved. Settings are only discarded when they cannot be recognized at all (written by a different project, or corrupted) — in that case the original file is backed up under <plugin folder>/settings-backups/ before defaults are used.

Rolling back

Every install snapshots the plugin folder it is about to overwrite, so rollback is one command:

npm run restore:vault -- "<vault>/.obsidian/qnalog-install-backups/<timestamp>" "/path/to/your/vault"

restore:vault reads the plugin id and version from the backup's manifest.json, restores that folder, and snapshots your current folder first — so the rollback itself is undoable (the undo command is printed). It rejects ids that are not safe single directory names and refuses linked or overlapping restore trees before writing. It also tells you which plugin id Obsidian currently has enabled; add --set-enabled to rewrite community-plugins.json instead of switching in the UI. Pass the vault path explicitly when the backup lives outside a vault.

Other development and check commands: see CONTRIBUTING.md.

License & credits

MIT — see LICENSE, NOTICE, and the maintenance policy in MAINTAINING.md.

Copyright (c) 2026 Lynnx (original LexVoice work); modifications copyright (c) 2026 Q&A Log Team.