memsearch¶

Cross-platform semantic memory for AI coding agents.

Install the plugin, get persistent memory. No commands to learn, no manual saving -- your agent remembers what you worked on across sessions, automatically.

For Agent Users¶

Pick your platform, install the plugin, and you're done. memsearch captures conversations, indexes them with hybrid search, and recalls relevant context when your agent needs it.

Claude Code Plugin¶

The most mature plugin. Marketplace install, zero config.

# In Claude Code:
/plugin marketplace add zilliztech/memsearch
/plugin install memsearch

Shell hooks + SKILL.md with context: fork subagent. Conversations are auto-summarized via Haiku and recalled via semantic search -- all without polluting your main context window.

Claude Code Plugin docs Troubleshooting

OpenClaw Plugin¶

Native TypeScript plugin with kind: memory. Per-workspace memory isolation out of the box.

openclaw plugins install clawhub:memsearch
openclaw gateway restart

Three tools (memory_search, memory_get, memory_transcript) with per-workspace isolation. Point an agent's workspace at a project directory to share memories with other platforms automatically.

OpenClaw Plugin docs

OpenCode Plugin¶

TypeScript plugin with daemon-based capture from OpenCode's SQLite database.

bash memsearch/plugins/opencode/install.sh

OpenCode Plugin docs

Codex CLI Plugin¶

Shell hooks, similar architecture to Claude Code. Requires --yolo mode.

bash memsearch/plugins/codex/scripts/install.sh
codex --yolo

Codex CLI Plugin docs

One Memory, All Platforms¶

All platforms share the same markdown memory format and derive collection names from the project directory using the same algorithm. A conversation in one agent becomes searchable context in all others -- no extra setup needed.

	Claude Code	OpenClaw	OpenCode	Codex CLI
Plugin type	Shell hooks	TS plugin	TS plugin	Shell hooks
Capture	Stop hook + Haiku	agent_end hook	SQLite daemon	Stop hook + Codex
Recall	SKILL.md (fork)	memory_search tool	memory_search tool	SKILL.md
Install	Plugin marketplace	`openclaw plugins install`	npm + opencode.json	`install.sh`

Platform comparison

For Agent Developers¶

Build custom agent integrations with the memsearch CLI or Python API.

Python API¶

from memsearch import MemSearch

mem = MemSearch(paths=["./memory"])

await mem.index()                                      # index markdown files
results = await mem.search("Redis config", top_k=3)    # semantic search
print(results[0]["content"], results[0]["score"])       # content + similarity

Python API reference

CLI¶

memsearch search "how to configure Redis?" --top-k 5
memsearch expand <chunk_hash>              # full section context
memsearch index ./memory/                  # index markdown files
memsearch watch ./memory/                  # live sync

CLI reference

Install¶

# Install the CLI (with ONNX embedding -- no API key needed)
uv tool install "memsearch[onnx]"

# Or with pip
pip install "memsearch[onnx]"

Getting Started

How It Works¶

graph LR
    subgraph "Source of Truth"
        MD["Markdown files<br/>.memsearch/memory/*.md"]
    end

    subgraph "Processing"
        SC[Scanner] --> CH[Chunker] --> EM[Embedder]
    end

    subgraph "Storage"
        MIL[(Milvus<br/>hybrid index)]
    end

    subgraph "Search"
        HS["Dense + BM25 + RRF"] --> RES[Top-K results]
    end

    MD --> SC
    EM --> MIL
    MIL --> HS

    style MD fill:#2a3a5c,stroke:#e0976b,color:#a8b2c1
    style MIL fill:#2a3a5c,stroke:#6ba3d6,color:#a8b2c1

Markdown is the source of truth. The vector store is just a derived index -- rebuildable anytime from the original .md files. Human-readable, git-friendly, zero vendor lock-in.

Core Features¶

Hybrid search -- dense vector + BM25 sparse + RRF reranking for best recall
Smart dedup -- SHA-256 content hashing means unchanged content is never re-embedded
Live sync -- file watcher auto-indexes on changes, deletes stale chunks
Memory compact -- LLM-powered summarization compresses old memories
Three-layer progressive disclosure -- search summaries, expand to full sections, drill into original transcripts
Cross-platform memory sharing -- shared Milvus backend lets agents on different platforms access the same memories

Architecture Design Philosophy

Embedding Providers¶

Provider	Install	Default Model
ONNX (plugin default)	`memsearch[onnx]`	`bge-m3-onnx-int8` (CPU, no API key)
OpenAI	`memsearch` (included)	`text-embedding-3-small`
Google	`memsearch[google]`	`gemini-embedding-001`
Voyage	`memsearch[voyage]`	`voyage-3-lite`
Jina	`memsearch[jina]`	`jina-embeddings-v4`
Mistral	`memsearch[mistral]`	`mistral-embed`
Ollama	`memsearch[ollama]`	`nomic-embed-text`
Local	`memsearch[local]`	`all-MiniLM-L6-v2`

Milvus Backend¶

Mode	`milvus_uri`	Best for
Milvus Lite (default)	`~/.memsearch/milvus.db`	Personal use, dev -- zero config
Milvus Server	`http://localhost:19530`	Multi-agent, team environments
Zilliz Cloud	`https://in03-xxx.zillizcloud.com`	Production, fully managed -- free tier

License¶

MIT