Dictate your trading thoughts each day and let an agent file them as dated notes

How it works

1. Make one folder for daily notes, one file per trading day named by date. A plain-text or Obsidian folder works; keep it on your own machine or in storage you control.
2. After the session, open any voice-to-text tool (your phone's dictation or a desktop app) and talk for two to five minutes: what you noticed, what you hesitated on and why, what you passed on and why, and how you felt. Don't edit while you talk.
3. Paste the raw transcript into the agent with the filing prompt. It puts your words under fixed headings (market observations, decisions and reasons, passed on, state and mood, questions for later), fixes obvious dictation errors, and leaves anything it can't place under "unsorted".
4. The agent shows you the filed note and a list of every edit it made. You correct it. Only after you confirm do you save the note under today's date and keep the raw transcript unchanged beside it, by pasting both into the folder. An ordinary chat window can't write to your disk; let the agent write the files only if that session already has access to the folder.
5. Point your other agents (weekly review, playbook interview, research wiki) at the folder as read-only context, with a standing instruction to quote the note and date whenever they rely on it.
6. Once a month, ask the agent to list the questions you left for yourself that you still haven't answered.

Copy-paste prompt

You file my dictated end-of-day trading notes. You do not give trading advice, you do not add opinions, facts or
market commentary of your own, and you never invent anything I did not say.

Raw transcript: {{transcript}}. Date: {{date}}. Recorded at: {{recorded_time}} ({{time_zone}}).

1. Put my own words under these headings: Market observations; Decisions and reasons; Passed on, and why;
   State and mood; Questions for later. Keep my wording. You may only fix obvious dictation errors and drop filler words.
2. Anything you can't place goes under "Unsorted", word for word.
3. If I name a specific security, replace it with "[name removed]" unless {{keep_names}} is yes.
4. Show me the filed note, then a list of every dictation fix you made (before and after). Do not save anything.
5. Only after I reply "save", output two complete fenced blocks for me to paste into {{notes_folder}}: the filed
   note, labelled {{date}}.md, and the raw transcript unchanged, labelled {{date}}-raw.txt. Do not claim you saved
   them. Write the files yourself only if this session already has confirmed write access to {{notes_folder}}.

Guardrails

• The agent only rearranges your words. It adds no commentary, facts or advice, and lists every edit it made.
• The raw transcript is kept unchanged next to the filed note, so you can always check what you actually said.
• Nothing is saved until you have read the note and confirmed.
• Other agents read the folder read-only and must quote the note and date they rely on.
• These are private records. Check your dictation tool's and model provider's data terms before you dictate anything sensitive.

What can go wrong

Dictation mishears trading vocabulary, numbers and names, and an agent asked to "clean up" will smooth a misheard number into something plausible. That's why every fix is listed and the raw transcript is kept. The slower failure is drift into summaries: let the agent paraphrase, and after a few months the folder holds its words, not yours, and every agent reading it learns its view of you. Keep the "your wording only" rule strict. Note the recording time too; a note made hours after the close is a different record from one made right after it.

Source and references