aru-labs/lossless-memory is a long-term memory repository that preserves conversation records without summarizing them, for personal AI. The September 23, 2026 commit 7b1100a rewrote the English FAQ, finished the Japanese FAQ, and added Shal as a co-author. According to the commit message, answers 1~11 of the FAQ were reviewed by Shal and two outside readers. The author credits in README, LICENSE, and pyproject.toml changed from Aru & Cece to Aru & Cece & Shal.
The revised docs/faq.md explains that this repository is a memory system for one specific person. It preserves records in their original form, uses SQLite, and envisions connecting to commercial or local AI by swapping out the interface module. The README presents as supported input paths a generic {ts, role, text} JSONL format and a format that reads conversation files from specific chat apps. It clarifies that other apps require separate importers, but the memory system itself does not change. The FAQ also contains an explanation that this repository is a memory component of a larger system, and an example of combining it with voice cloning technology.
The README announces that development is paused for the time being due to personal financial circumstances. It also states that there are no benchmarks, and draws a line that the figures listed in the documentation are operational measurements, not comparisons with other systems. This change is an update that tidied up the FAQ, the project description, and the author credits, and it presents no material supporting performance comparisons.