# GitHub - aru-labs/lossless-memory: Lossless long-term memory for a personal AI

> Personal AI keeps every conversation line with timestamps, no summaries.
- Title: GitHub - aru-labs/lossless-memory: Lossless long-term memory…
- Summary: Personal AI keeps every conversation line with timestamps, no summaries. A personal AI system stores raw conversation logs with timestamps to avoid…
- Keywords: lossless-memory, personal-ai, timestamped-conversations, temporal-backbone, lll-index, jsonl-logs, sqlite-fts5, semantic-search, exact-search, single-user-system, english-japanese, time-queries
- Source: GitHub — https://github.com/aru-labs/lossless-memory
- Read time: 1 min
- Topics: lossless-memory, personal-ai, timestamped-conversations, temporal-backbone, lll-index, jsonl-logs, sqlite-fts5, semantic-search, exact-search, single-user-system, english-japanese, time-queries
## What it does
Stores raw conversation logs without summarization.

> Never summarize, keep every line, put a timestamp on everything
## Why it matters
Prevents forgetting what was actually said and when.

> The thing that matters most to a person who talks to the same ai every day: what was actually said, and when
## How it works
Timestamps every record and searches by time first.

> Search by time first, words second
## The Temporal Backbone
Uses time as the primary search axis.

> Time is not metadata here; it is the primary axis
## LLL index
Tracks current topic with human-written markers.

> The ai reads it; the human writes it
## Real operation
Has run since June 2026 for one user.

> This implementation has been running every day since 2026-07
## Design pillars
No summaries, raw logs, and time-based search.

> It is not a summarizer. There is deliberately no summarization step anywhere in the pipeline
## Why it exists
Fixes ai forgetting conversations over time.

> This was built for one person who has talked to ai assistants every day for years and watched each of them forget
## How it scales
Reindexes every 10 minutes without downtime.

> A small daemon reindexes incrementally on a fixed interval (default: every 10 minutes)
## Limitations
Only works for one person and one machine.

> Designed for one person and one ai, running on one machine. No server, no cloud
## What it's not
Not a vector database or benchmark system.

> It is not a vector database wrapper. Semantic search is the last resort here, not the first
## Key takeaway

A personal AI system stores raw conversation logs with timestamps to avoid forgetting details, working since June 2026.