+- memory

+- data

+an unstructured data logger that accepts anything — text, voice, images, sensor readings, location, whatever you throw at it — and uses an LLM to generate dynamic schemas on the fly rather than requiring you to pick a category or fill a form. the problem with most logging tools is they impose structure upfront: here's a field for "mood," here's one for "food eaten," here's one for "notes." the structure forces you to fit your life into its buckets. universal data capturer inverts this: log freely, and let the system figure out what you're tracking and how to structure it. over time, the schema adapts to your actual behavior — if you keep mentioning sleep quality, it starts tracking that specifically.

+

+the technical core is an LLM that ingests each new entry, compares it to what's been logged before, and decides: does this fit an existing schema? should I update the schema? is this a new category entirely? this is surprisingly tractable with a modern LLM — the hard part is making the schema evolution feel stable rather than constantly shifting. the user-facing value is that it lowers the activation energy for logging to near zero: just dump whatever is on your mind, and the system handles the rest. the data becomes queryable and analyzable later because the schema inference has done the structuring work.

+

+connects to [[axon|axon]] which is the personal memory layer that this feeds into — universal data capturer is the ingestion side of axon's retrieval layer. [[life-search|life search]] depends on having rich logged data to search over. [[sensor-capturer|sensor capturer]] is the hardware version of the same capture impulse — physical sensors instead of text/voice. [[me-model|me model]] would be trained on the data this produces. [[episodic-memory-builder|episodic memory builder]] is a more focused version for a specific type of log entry. [[ifttt-personal|personal IFTTT]] could sit on top to trigger automations based on patterns the capturer detects.