BAI-LAB
lab · ChinaScores
1 product on the map — 1 open.
Openness
5 high confidence- license
- Apache-2.0 (library, MCP server, ChromaDB variant and playground)
- source
- public
- core features withheld
- no — the 73-path tree at 587ed77 carries one root Apache-2.0 LICENSE over the library, MCP, ChromaDB, playground and evaluation directories, with no ee/, enterprise/, commercial/ or proprietary/ path; an academic lab's release
MemoryOS is published under Apache 2.0, and every variant, the library, MCP server, ChromaDB build and playground, runs from the published source against the user's own model keys. It is a university lab's research release, and its hosted platform is an invitation-only demo of the same open playground.
- https://baijia.online/memoryos/ recorded 2026-11-20
The project site gives the contact address baiting@bupt.edu.cn and links the GitHub repository and the research paper.
- https://raw.githubusercontent.com/BAI-LAB/MemoryOS/main/LICENSE recorded 2026-11-20
The LICENSE is the Apache License 2.0 text with the appendix line "Copyright [2025] [BAITING]".
- https://raw.githubusercontent.com/BAI-LAB/MemoryOS/main/README.md recorded 2026-11-20
The README carries an Apache 2.0 badge, lists PyPI, MCP, ChromaDB, Docker and playground set-ups, and announces "Open-sourced the Playground platform"; the hosted MemoryOS Platform asks for an invitation code.
Adoption
1 low confidenceAdoption is read from PyPI downloads of memoryos-pro, the package the README installs; the similarly named memoryos package belongs to MemTensor's MemOS and is not counted. Installs from source, the MCP server and the Docker set-up do not appear in this figure.
- https://pypi.org/pypi/memoryos-pro/json recorded 2026-11-20
memoryos-pro has a single 0.1.0 release from 2025-07-09 whose metadata still carries the packaging template's placeholder author and project links.
- https://pypi.org/pypi/memoryos/json recorded 2026-11-20
The memoryos package on PyPI is "Intelligence Begins with Memory" by MemTensor, repository MemTensor/MemOS.
- https://pypistats.org/api/packages/memoryos-pro/recent recorded 2026-11-20
89 downloads of memoryos-pro in the last month.
Capability
5 medium confidenceMemoryOS covers the whole cycle: it summarizes dialogue into segments, promotes the ones a user keeps returning to into a profile and knowledge base, and lets cold segments cool and drop out as capacity fills. Its forgetting is driven by access and recency rather than by any check that a fact is still true.
- https://raw.githubusercontent.com/BAI-LAB/MemoryOS/main/memoryos-pypi/mid_term.py recorded 2026-11-20
Segment heat adds a recency term from compute_time_decay with a 24-hour tau, and evict_lfu removes the session with the lowest access frequency when capacity is reached.
- https://raw.githubusercontent.com/BAI-LAB/MemoryOS/main/memoryos-pypi/updater.py recorded 2026-11-20
process_short_term_to_mid_term takes the QA pairs evicted from short-term memory, asks the model whether each continues the previous page, and summarizes them into mid-term memory.
- https://raw.githubusercontent.com/BAI-LAB/MemoryOS/main/README.md recorded 2026-11-20
Short-term dialogue is consolidated "into meaningful segments" in mid-term memory; segments accumulate "heat" and, past a threshold, "User profile insights are extracted" and facts added to user and assistant knowledge; the retriever reads all layers.