tigerless-labs/agent-memory is a long-term agent memory runtime shared by tools such as Claude Code and Codex CLI. The version badge is 0.1.0, it requires Python 3.12 or later, and it does not depend on any API key. Tools that can run shell commands, such as Muse Code, share the same repository. Markdown files in a single repository serve as the source of memory, while the SQLite index is treated as a disposable cache.
This structure is designed so that agents do not load the whole memory into context every time but instead locate and read the files they need. Retrieval first returns an L0 list containing a one-line summary, file path, anchor, and score instead of the memory body. An anchor points to the location of the memory within the file. Agents open an outline or the full content as needed. Read paths are divided into deterministic injection via MEMORY.md, FTS5-based BM25 search, and the use of ls and grep in a plain directory. If the vector plugin is enabled, search results are fused with RRF. If one path fails to find the desired memory, the others can be used. Writes happen automatically at conversation boundaries, and a separate sleep-time consolidation process merges or discards memories on its own clock.
The repository is MIT licensed and the primary language is Python. The provided description contains no measurements comparing retrieval quality or performance. So while the three paths and the file-centric storage approach can be verified, there is not enough evidence to conclude that retrieval is better than that of other long-term memory systems.