← Compare Compare · per product
Hydrate vs MemSearch (Zilliz)
personal memoryZilliz's markdown-source-of-truth + Milvus memory layer for AI agents. The closest in the field to a git-shareable model — but 'commit your .md files' is not canon propagation with attribution.
What MemSearch (Zilliz) is
| Stack / runtime | Markdown source of truth + Milvus (Lite/Server/Zilliz Cloud) |
| Licence | Open source (Zilliz-backed) |
| Storage | Markdown .md files indexed in Milvus |
| Hooks | Automatic conversation capture |
| Injection | Automatic semantic recall |
| Scope | Cross-agent: Claude Code, Codex, OpenCode |
| Maturity | Vendor-backed (Zilliz/Milvus) |
| Price | Free OSS; Zilliz Cloud for production |
| Version referenced | Open-sourced 2026 |
| Source | github.com/zilliztech/memsearch |
Where MemSearch (Zilliz) leads
- Markdown as source of truth (human-readable, editable, version-controllable)
- Pluggable embeddings (OpenAI/Gemini/Voyage/Ollama/local; default ONNX bge-m3, no key)
- Hybrid dense + BM25 + RRF retrieval today
- Milvus scaling path (local -> server -> cloud); open source; cross-agent
Where Hydrate leads
- Decision capture, not just semantic recall of conversation
- Canon propagation with attribution + merge (not just 'version-control your md')
- Live cross-vendor sessions + orchestration on the same substrate
- Single dependency-free binary (no Milvus, no embedding-model download)
Verification notes
Most credible cross-agent claim in the personal-memory field. Differentiator vs Hydrate must be stated precisely: canon propagation + attribution + LIVE bridge, not files-in-git. Verify whether MemSearch has any attribution/merge model for teams.
Sources
- Primary repo + docs https://github.com/zilliztech/memsearch