The ranking covers 44 qualifying repositories out of 57 candidate GitHub pages fetched directly on 2026-11-12 (3 requested pages failed; the rest fell outside scope — orchestration frameworks, caches, research agents or example-only repos). Stars are an exact snapshot from each repository's own page, grouped into memory-first systems, shared/multi-agent memory, and company/team knowledge brains. This is a researched landscape, not a provably exhaustive universe.
| # | Repository | Stars | Category | Capability | License | Source |
|---|
Inclusion: a repository qualifies if its own implementation provides persistent memory reusable across AI-agent sessions or tools, memory shared among agents or teams, or a durable personal/team/company knowledge brain agents can directly use. Company-brain entries may be RAG, knowledge-graph or document systems when agents can consume the resulting organizational context.
Exclusion: generic orchestration frameworks where memory is incidental (e.g. CrewAI, MetaGPT, Dify, Haystack, LlamaIndex), semantic caches (GPTCache), research agents (STORM, gpt-researcher), directories and awesome lists, and repositories holding only examples or integrations for a closed product (getzep/zep examples repo; the hosted Zep memory engine itself is not in this open-source ranking).
Caveats: stars were read directly from each GitHub page on 2026-11-12 and are a popularity snapshot, not a quality score. dnotitia/akb is flagged source-available (BSL 1.1). A broader corroboration sweep of 887 extracted web mentions informed discovery but never overrode the direct counts.
Direct GitHub repository-page fetches, 2026-11-12: 57 candidates fetched, 3 pages failed, 44 met the strict scope above and are ranked here; stars, descriptions and licenses taken verbatim from each page. Category and one-line capability assigned editorially from each repo's own description. Out-of-scope candidates listed in methodology were cut from chart and table.