Import external conversations

Thinking you did elsewhere - with another LLM, a colleague, or in your own notes - can count toward your competency map. Fill the template, import it, and each entry becomes a learning signal on its module. The conversation text itself is never stored. Back to the mentor.

The template

Fill this template with conversations you had outside the app - another LLM, a colleague, your own notes. Each entry needs a source: line, and a module: address or tags: so it can be mapped. What is recorded per entry is exactly three things: the module it maps to, your source attribution, and an anonymous re-import id. The conversation text is never stored unless you turn import storage on - the free zone below the header lines exists to separate entries and to remind you what each one was.

# External conversations - import file
#
# How to fill this in:
#   1. Duplicate the "## entry" block below, once per conversation.
#   2. Replace every <placeholder> - an entry with an unreplaced
#      placeholder is refused rather than recorded wrong.
#   3. source: is required - which LLM or which person it was.
#   4. Keep module: (a P<pillar>.<track>.<nn> address) OR delete that
#      line and use tags: (comma-separated topics) instead.
#   5. Everything below the header lines is the free zone - paste the
#      conversation or your notes there. It is never stored; it only
#      separates entries and helps you remember what this was.

## entry
source: <which LLM or person - e.g. ChatGPT, or: colleague - Sam>
date: <YYYY-MM-DD - optional, delete if unwanted>
module: <P10.attention.01 - or delete this line and use tags:>
tags: <attention, embeddings - comma-separated, optional>

<The free zone. Paste the conversation or write your notes here.
Nothing in this zone is stored or sent anywhere - what is recorded
is only: the module it maps to, and who the conversation was with.>

## entry
source: ChatGPT
date: 2026-08-15
module: P10.attention.01

A filled example: we discussed why attention needs a separate value
projection, and where the residual stream fits. Delete this entry or
replace it with your own.

Import

A file is read the same way a paste is (txt, md, PDF, DOCX). Re-importing the same entries is safe - they are skipped, never duplicated.

Import conversations — TransformerLab