Skip to content

Use Transcripts as Context in Codex and Claude

Ideas spoken on a walk and thoughts recorded as they come to mind at work can become context for your next AI task. Ayumi saves transcripts as Markdown files, so you can ask a desktop AI app to read the folder where those records live.

Instead of pasting each transcript into a chat, use your growing collection of notes as a starting point. Ask the AI to consult those files when organizing this week’s ideas or drafting a plan.

Record on iPhone or Apple Watch
↓ Ayumi transcribes and saves
Journal folder in iCloud Drive (Markdown)
↓ iCloud syncs
The same folder downloaded on your Mac
↓ Read through a Codex project or Claude's connected folder
Work with AI on a reflection, proposal, or implementation plan
↑ Continue from mobile using the AI app's supported remote features

Ayumi handles recording, transcription, and file storage. iCloud Drive handles sync. Configure file access and remote connections in your AI app. No Ayumi-specific API or plugin is required.

1. Save to iCloud Drive and open on your Mac

Section titled “1. Save to iCloud Drive and open on your Mac”
  1. Choose a folder in iCloud Drive as Ayumi’s journal location. This guide uses AyumiJournal as an example name.
  2. Record a short note on your iPhone and confirm transcription has finished and the entry has been saved. For Apple Watch recordings, wait for transfer to the iPhone and transcription to finish.
  3. On a Mac using iCloud Drive with the same Apple Account, open that folder in Finder.
  4. Open the relevant .md file and check that you can read the transcript. Download any files that are still cloud placeholders.

Transcripts are saved as journal entries, so you do not need a separate export for AI use. See Markdown Format and Sync with iCloud for details.

Open the synced AyumiJournal folder as a project in the desktop app. Start by asking it to read one specific entry. To use notes alongside a development project, point to the journal location and grant the requested file access. The OpenAI desktop app guide explains choosing a folder or project as your workspace.

This folder contains Ayumi transcripts.
Organize product ideas from entries created in the last seven days.
Use created_at in the frontmatter to select dates, and cite the source file for each idea.
Label anything inferred rather than recorded. Leave the original files unchanged.

Connect AyumiJournal in Claude Desktop and start a task that reads it. For example, ask for the background and open questions for a proposal based on your notes.

Cowork currently runs in the cloud by default and reaches permitted files on your Mac through the desktop app. See Claude’s architecture overview.

Cowork and Claude Code are separate features. Cowork can help with documents and research; Claude Code desktop supports development work. You can also give Claude Code access to transcripts as background for development. See the Claude guide.

3. Continue from mobile or another desktop

Section titled “3. Continue from mobile or another desktop”

When you can reach the Mac that holds your synced notes, you can continue working away from it. iCloud file sync and your AI app’s remote connection each need their own setup.

AppHow to continueConditions for reading local files
CodexPair your device in desktop connection settings, then connect through Codex in ChatGPT mobile (Remote in older interfaces) or another supported desktopHost app open; Mac awake and online; same account and workspace; required file permissions
Claude CoworkOpen a session started on desktop in Claude mobile or webClaude Desktop open and online; access limited to connected folders

Codex uses the connected host’s projects, files, and permissions. Follow OpenAI’s Remote connections guide for setup.

Cowork sessions in projects tied to local folders must start on desktop. Uploaded project knowledge and a connected folder on your Mac have different access rules. Check Claude’s guide to supported surfaces.

What you recordWhat you can work on with AI
Feature ideas spoken on a walkGroup related ideas and identify requirements and questions before implementation
Notes about thoughts that come to mind at workOrganize ideas, concerns, and things to try next
Thoughts recorded while researchingSeparate findings from hypotheses and identify what to investigate next
Daily voice journal entriesReflect on recurring interests and concerns

Start with reading files and an answer in chat. When you want to save a result, choose an output folder separate from the original records so you can review both later.

Read the transcripts of thoughts I recorded as they came to mind at work.
Organize ideas, concerns, and things to try next.
If background or conditions are not recorded, mark them as unconfirmed rather than guessing.
Cite the source file and leave the original unchanged.
Show the draft in this chat first.

Content read by a hosted AI is sent to its model provider for processing. Local-first storage does not mean AI processing stays on your device. Local files opened by a Cowork cloud session are processed on Anthropic’s servers. See Claude’s architecture overview.

To keep access narrow, connect a dedicated folder containing only the records you want the AI to read. You do not need to include recordings or entries with location metadata; you can copy just the relevant text.

Alongside asking the AI to leave originals unchanged, use read-only permissions if the app offers them. Review generated material before publishing or sending it, and follow your AI service’s data settings and your organization’s rules.

ProblemWhat to check
The latest transcript is missingEntry saved in Ayumi; iCloud sync complete; .md body readable on the Mac
The folder is visible but the AI cannot read itFile access granted to that folder in the AI app
Local files are unavailable from mobileMac and desktop app online; remote connection configured; Cowork session started on desktop
Yesterday’s entry is not in the expected date folderSelect using created_at and its time zone rather than folder names alone

From the same maker: Kitae, a dead-simple strength-training log

Follow Ayumi: X (@AyumiJournalApp)