● Official
zilliztech
⚡ Instant
Memsearch
by zilliztech · zilliztech/memsearch
Unified semantic memory across Claude Code, OpenClaw, OpenCode, Codex — Markdown source files, Milvus-backed retrieval.
Memsearch is Zilliz's plugin/skill for persistent agent memory. Files are plain Markdown so you can read/edit them; Milvus provides the semantic-search layer. Works across multiple agent harnesses.
Why use it
Key features
- Markdown source of truth
- Milvus-backed vector search
- Multi-harness (Claude Code, OpenClaw, OpenCode, Codex)
- CLI + Python API
- Cross-session memory
Live Demo
What it looks like in practice
ready
Use Cases
Real-world ways to use Memsearch
When to use: You use Claude Code AND Codex and want a single memory.
Flow
-
Install
/plugin install memsearch (or pip).✓ Copied
→ Plugin active
-
Init
memsearch config init.✓ Copied
→ Config saved
-
Use
remember/recall from any agent harness.✓ Copied
→ Shared memory
Outcome: Memory persists across agents — no re-explaining.
Combinations
Pair with other MCPs for X10 leverage
Use Ars Contexta's architecture as memsearch sources
Combine memsearch-skill with arscontexta: Use Ars Contexta's architecture as memsearch sources✓ Copied
Cost & Limits
What this costs to run
- API quota
- Local + Milvus
- Tokens per call
- Recall is small
- Monetary
- Free OSS
- Tip
- Run Milvus in Docker locally; cloud is overkill for personal use
Security
Permissions, secrets, blast radius
Minimum scopes: fs-read fs-write
Credential storage: Markdown + Milvus DB
Data egress: Local Milvus by default
Troubleshooting
Common errors and fixes
Recall misses
Reindex; or tighten chunk size
Alternatives
Memsearch vs others
| Alternative | When to use it instead | Tradeoff |
|---|
| stash-mcp | You want a tiny Go binary | No Milvus scale |