Lifelog
Your real-life conversations, meetings, and texts, captured into one place you can ask questions of.
The problem
The record of your own life is scattered across a pendant, a notes app, and your phone, and none of it is searchable.
What you said, what you agreed to, what someone told you: it lives in three places that do not talk to each other. A Limitless pendant hears the room, Granola takes the meeting notes, iMessage holds the texts, and each one is a separate silo you cannot query. By the time you need the detail, you are guessing from memory. A write-only pile of transcripts is not a memory; it is just storage.
What it does
capture, group, askThree sources, one store.
Every fifteen minutes it pulls Limitless pendant audio, Granola meeting notes, and iMessage into one SQLite file. Writes are idempotent, gaps get backfilled, so a day is never silently empty. A local Mac agent pushes the iMessage side.
Raw logs become conversations.
An LLM segments the day's fragments into real conversations, extracts the commitments, and rolls them up into hourly, daily, and weekly briefs. The unit of meaning is the conversation, not the log line.
Ask Claude about your own life.
An MCPB bundle drops the server into Claude Desktop. Tools like search_conversations and get_commitments answer from the record, so "what did I commit to this week?" returns a fact, not a guess.
A dashboard for the daily review.
A dense, quiet surface to skim what got captured, spot a broken ingest, and open the one conversation worth reading. Names come from your Google Contacts, so iMessage shows people, not raw phone numbers.
One question in, one answer out
How it's built
A headless pipeline with a quiet dashboard bolted to the visible end.
A Typer CLI and a FastMCP server share one Peewee-backed SQLite file, WAL mode, with FTS5 tables for full-text search. A background scheduler ingests every fifteen minutes; a process-wide lock keeps it from overlapping with agent-triggered ingests. The hard part is not the tools, it is trusting the data: idempotent writes, gap backfill, and stale-date rechecks, so the record is complete before anyone asks it a question.
This is what six weeks looks like
We built Lifelog for ourselves. We can build yours.
Every product in our labs was designed and shipped by the same two-person team that would build your MVP. If you have an idea worth building, that's the whole pitch.
Start a build →