Short answer: open Logseq on today's Journal page, click a bullet, press ⌃+⌥+R (Mac) or Ctrl+Alt+R (Windows/Linux), speak, press again. AICHE inserts prose into that block. Add [[page links]] and ((block refs)) by hand after capture. Files stay local Markdown on disk.
Logseq is an outliner: the Journal, atomic blocks, page links with [[...]], and block references with ((...)) only pay off when blocks contain enough context to embed elsewhere. Voice fills the block; you type the graph syntax in a quick review pass.
Typed blocks abbreviate ("API redesign delayed Q2."). Spoken blocks carry owner, constraint, and date ("Decided to push REST-to-GraphQL to Q2; Chen owns design review; frontend blocked on flexible queries."). The second block survives a ((block ref)) six months later.
How AICHE Fits Into Logseq
- Open Logseq. It lands on today's Journal page by default.
- Click into a bullet, or click below the last block to start a new one.
- Press ⌃+⌥+R on Mac or Ctrl+Alt+R on Windows/Linux to start recording.
- Speak the block. One thought per recording works best.
- Press the hotkey again. AICHE transcribes, cleans filler, and inserts the text into the active bullet.
- Press Enter for the next block. Press Tab to indent it as a child.
AICHE does not type Logseq syntax for you. It writes plain prose into the cursor. Page links, block references, tags, properties, and TODO/LATER/NOW markers are still added by hand. That is the right division of labor: speak the content, type the structure.
The Journal Page As Capture Funnel
The Journal is the lowest-friction surface in Logseq. It opens automatically, it is already dated, and nobody expects it to be organized. That makes it the obvious target for voice.
Morning dump. Open Logseq, hotkey, speak for sixty seconds about what is on your plate today. Stop. You now have one block of dense context. Press Enter, then either keep speaking the next item or break the first block at the line returns AICHE inserted. By the end you have eight or ten bullets that would have taken ten minutes to type.
Drive-by capture. The reason ideas die is that getting them into a tool takes longer than the idea itself. With the hotkey global, you can be in any window, switch to Logseq with a keyboard shortcut, click a bullet, speak for ten seconds, and be back in the previous app inside thirty seconds. The cognitive tax stays under what the thought is worth.
Weekly review. At the end of the week the Journal pages contain raw blocks. Scan them, add [[page links]] to the ones worth promoting, tag them, and let Logseq's backlinks pull them into their proper homes. Voice keeps the funnel full; the keyboard still does the structure pass.
Block References Reward Complete Blocks
Logseq's signature move is ((block-id)): drop a reference and the actual content of that block renders inline on the new page. The trick is that the embed only makes sense if the source block was written to stand alone.
Typed blocks fail this test constantly. You abbreviate because typing more feels expensive, then six weeks later you embed the block on a project page and it reads like a fragment. Spoken blocks default to context. When you talk, you include the why, the who, and the constraint without thinking about it.
Speak the block once, write it complete, and every future ((reference)) lands a usable unit. The graph gets denser without you doing extra work later.
Meeting Notes Without Typing Through The Meeting
During a call, frantic typing splits your attention and produces sparse bullets nobody can act on. Keep Logseq open on the Journal. When something concrete happens, hit the hotkey and speak one block: "Decision to ship behind a flag for v2.8, Olu owns the rollout plan, target review on the 28th." Hotkey to stop. Back to listening.
An hour-long meeting produces ten to fifteen of these, each with the decision, the owner, and the next step. After the meeting, walk through, add TODO markers to the action items, tag the relevant project pages, link the participants. The capture pass and the structure pass are now separate jobs, and the capture pass barely took attention away from the conversation.
Reading And Literature Notes
Keep Logseq beside whatever you are reading. When a passage triggers a reaction, switch over, hotkey, speak the reaction in full: the author's claim, your pushback, the cross-reference to something else you read. Fifteen seconds of speech beats sixty seconds of typing and does not pull you out of the text for as long.
After the session, those blocks are already complete enough to live in your graph. Add [[author]] and [[book title]] links, drop a #literature-notes tag, and you have something queryable.
Datalog Queries Reward Detail
Logseq's query system is one of the most powerful features in the tool, and it punishes lazy writing. A query that surfaces "every block tagged with #decision from the last 30 days where the owner is [[Chen]]" only returns useful rows if those blocks contain the words and links it needs.
Voice helps because spoken blocks are denser by default. You name the people, the dates, the systems, the reasons. The query in March 2027 finds the block you spoke today because the actual content of the block is specific, not a four-word summary that made sense to you at the time.
Local Graph, Cloud Transcription
Logseq stores your graph as Markdown or Org-mode files in a folder you chose. No proprietary database. No required sync account. You can grep your notes with ripgrep if you want.
AICHE is the voice-to-text layer in front of that. Audio is streamed for cloud transcription, processed, and discarded immediately after processing, within 1 second. No persistent audio copy. The text lands in the local file Logseq is already managing. Nothing new gets added to your data chain beyond transcription - no extra cloud database holds your notes, no audio archive exists to leak later. See Data security and privacy for the full picture.
If you are working somewhere without network, Offline Voice-to-Text Mode queues recordings until you reconnect, then drops the text into whichever block you point to. Useful on a plane, in a basement office, or during the kind of conference Wi-Fi that pretends to work.
Tips That Actually Matter In Logseq
One thought per recording, one recording per block. Logseq's whole model is that blocks are atomic. Match the recording boundary to the block boundary and the graph stays clean.
Speak prose, type the brackets. [[, ((, #, TODO, LATER, and property fields like due:: are punctuation against an outliner. Dictating those characters interrupts the flow and rarely produces the right output. Speak the meaning, add the syntax in the editing pass.
Turn on Content Organization for longer captures. Recapping a meeting or summarizing a chapter often runs past sixty seconds. With Content Organization on, AICHE returns the transcript with logical breaks you can split into separate blocks instead of one giant wall.
Use a code-aware profile for technical blocks. If you are dictating API names, library names, CLI flags, or identifiers, the Software Development profile (Pro) holds those tokens together instead of trying to spell them phonetically. See Smart Insert for how the inserted text lands in the active field.
Clean Language for published graphs. If you publish your Logseq graph or share it with a team, Clean Language strips the spoken quirks without rewriting your meaning. The block reads like prose someone wrote, not a transcript someone forgot to edit.
Multilingual thinking is fine. Think in German, Polish, Mandarin, or whatever your brain runs in. With translation enabled, the block lands in English while you keep the cognitive load on the idea, not the wording. See Language support.
Common Questions
Q: Does AICHE work in the Logseq desktop app on all three platforms?
A: Yes. AICHE inserts into whichever text field has the cursor. Logseq's editor is a standard text field on macOS, Windows, and Linux, so the hotkey and the insertion work the same way in all three.
Q: Will AICHE write [[page links]] or ((block refs))) for me if I say them out loud?
A: No, and you do not want it to. The bracket characters are awkward to dictate reliably and the linking decision usually wants a beat of thought. Speak the content, add links during your review pass. The block stays useful even without them.
Q: I keep my graph in Org-mode, not Markdown. Does that change anything?
A: No. AICHE inserts plain text at the cursor. Logseq still saves the file in whatever format you configured. Org-mode users get the same workflow as Markdown users.
Q: What about my Logseq plugins - any conflicts?
A: AICHE runs outside Logseq as a desktop voice layer, not a plugin. Plugins like the Smart Search, Bullet Threading, or Awesome Links extensions all keep working. AICHE just delivers text into the cursor position; whatever runs after Enter is Logseq's concern.
Q: Does it work with the mobile Logseq app?
A: This page is the desktop hotkey workflow for the Logseq desktop client on macOS, Windows, and Linux. AICHE also has mobile recording flows, but Logseq integration here is desktop-focused.
Q: I record in a quiet room but also during walks. Does the same hotkey behavior work?
A: Yes. Recording is a toggle - press once to start, press once to stop. You can pace around, talk through a problem out loud, then stop when you sit back down. The block in Logseq waits for you.
Q: How does this compare to using AICHE with Obsidian or Roam?
A: The mechanics are identical (hotkey, speak, stop, text lands at the cursor). The difference is structural: Logseq is outliner-first, so the natural unit is a block, not a note or a paragraph. See AICHE with Obsidian and AICHE with Roam Research for the equivalents.
Result: a Journal that fills with twenty to thirty detailed blocks a day instead of eight thin ones. Block references pull in usable context instead of fragments. Queries return rows worth reading. The graph compounds because the input throttle is no longer your fingers.
Try it now: open Logseq, land on today's Journal, click an empty bullet, press your hotkey, and dictate the last decision you made today - who, what, when, and why. Stop. You have your first complete block.