# braindump Capture a stream of loose thoughts verbatim, then break it into categorised llms files an agent can load — ideas, facts, questions, decisions — with every task pushed to your to-do list and every row cited back to the raw line. Say what is in your head and nothing else. `braindump` saves the text untouched first — before any interpretation — and only then parses it. Each parsed row points back to the exact line of the raw dump it came from, so a wrong classification loses nothing: the original is one hop away. ## What it does - **Capture first, parse second** — the raw dump is written to disk verbatim before a single word is classified. - **Categorised llms files** — the dump is split into `braindump__llms.md` files (ideas, facts, questions, decisions, tasks, people, references) with an `llms.txt` index over them, so an agent loads only the category it needs. - **Tasks go to the to-do list** — every actionable line is pushed to [todo](/skills/todo/) with its `path:line`, so nothing said in passing is lost. - **Meeting notes and voice memos** — the same pipeline turns a page of meeting notes into files that are usable as context the next time the subject comes up. - **Cited rows** — every line in every generated file ends with the raw path and line number it was parsed from, the same convention memory-central uses. Braindumps accumulate into a personal corpus: the categorised files are the long-term memory a future session loads, and the `llms.txt` index is how it finds the right one. **Use it for:** "braindump", "capture this", "get this out of my head", "note to self", a stream of loose thoughts you want kept.