maasv Memory

Provides structured long-term memory with semantic, keyword, and knowledge graph retrieval, entity extraction, temporal versioning, and experiential learning.

Security Scan

The skill's purpose (local long-term memory) is plausible and its runtime instructions mostly match that purpose, but the registry metadata omits the API keys and config requirements the SKILL.md tells you to create and the skill instructs modifying agent config and capturing conversations — these mismatches deserve attention before installing.

Purpose & Capability

The name/description (long-term, semantic/keyword/graph memory) aligns with the instructions to run a local maasv server and install an OpenClaw memory plugin. Requesting LLM and embedding providers is consistent with entity extraction and semantic search. However, the registry lists no required environment variables or primary credential while the SKILL.md clearly requires multiple API keys/configuration — a metadata mismatch.

Instruction Scope

SKILL.md gives explicit, bounded runtime steps: install the maasv Python server, fill a .env with provider API keys, run maasv-server, install the @maasv/openclaw-memory plugin, and edit ~/.openclaw/openclaw.json to point memory slot at the local server. The instructions will cause the agent to send conversation summaries to the local server (autoCapture) and to inject memories back into context (autoRecall). These behaviors are expected for a memory backend but you should be aware that conversation content will be written to a local SQLite DB and that external LLM/embedding providers (if used) will receive data.

Install Mechanism

This is an instruction-only skill (no install spec, no code files). The install steps are manual: pip install maasv[...], run maasv-server, and openclaw plugins install @maasv/openclaw-memory. Because nothing in the registry will be automatically downloaded or executed by the platform, install risk is limited to the third-party packages you choose to install yourself. Review maasv and the npm plugin source before running pip/npm installs.

Credentials

The SKILL.md requires multiple environment variables/API keys (MAASV_LLM_PROVIDER, MAASV_OPENAI_API_KEY, MAASV_ANTHROPIC_API_KEY, MAASV_EMBED_PROVIDER, MAASV_VOYAGE_API_KEY, optional MAASV_API_KEY) which are proportionate to the described functionality. The problem: the skill registry metadata lists no required env vars or primary credential, so the registry under-reports the secrets the user must supply. That mismatch makes it easy to overlook that you must provide potentially sensitive API keys or choose a fully-local option (e.g., Ollama) to avoid cloud calls.

Persistence & Privilege

always is false (good). The skill instructs editing the agent's config (~/.openclaw/openclaw.json) to register the memory plugin — this is normal for enabling a memory backend but is a persistent change to your agent setup. The maasv server will persist conversations and extracted entities to a local SQLite file; ensure you understand where that DB lives and how to secure or delete it (the SKILL.md mentions memory_forget for deletion).

What to consider before installing

This skill appears to be what it says (a local memory backend) but the registry metadata under-reports required credentials and configuration. Before installing: 1) Decide whether you will use cloud LLM/embedding providers (OpenAI/Anthropic/Voyage) — those require API keys and will cause data to leave your machine; prefer Ollama/local models for fully local operation. 2) Inspect the maasv GitHub and the @maasv/openclaw-memory npm package source to confirm you trust the maintainers. 3) Back up and secure the SQLite database the server creates and confirm where it will be stored. 4) Be aware the plugin will capture conversation summaries and inject memories back into agent context — test with non-sensitive data first and confirm delete/forget workflows work as you expect. 5) Because the registry omitted required env vars, a careful manual review of .env.example and any plugin docs is recommended before providing secrets.

Structured long-term memory for OpenClaw agents, powered by maasv.

Replaces the default memory backend with a cognition layer that includes 3-signal retrieval (semantic + keyword + knowledge graph), entity extraction, temporal versioning, and experiential learning.

maasv is entirely self-hosted. There is no maasv cloud service. You run the server on your own machine, and all data is stored in a SQLite file on your local disk that you own and control. Nothing is sent to maasv.

Install

This skill requires the @maasv/openclaw-memory plugin and a running maasv server.

1. Start the server

pip install "maasv[server,anthropic,voyage]"
cp server.env.example .env  # fill in API keys (see below)
maasv-server

2. Install the plugin

openclaw plugins install @maasv/openclaw-memory

3. Activate

// ~/.openclaw/openclaw.json
{
  plugins: {
    slots: { memory: "memory-maasv" },
    entries: {
      "memory-maasv": {
        enabled: true,
        config: {
          serverUrl: "http://127.0.0.1:18790",
          autoRecall: true,
          autoCapture: true,
          enableGraph: true
        }
      }
    }
  }
}

Required Credentials

The maasv server needs an LLM provider (for entity extraction) and an embedding provider (for semantic search). Configure these in your .env file:

Variable	Required	Purpose
MAASV_LLM_PROVIDER	Yes	anthropic or openai
MAASV_ANTHROPIC_API_KEY	If using Anthropic	LLM calls for entity extraction
MAASV_OPENAI_API_KEY	If using OpenAI	LLM calls for entity extraction
MAASV_EMBED_PROVIDER	Yes	voyage, openai, or ollama
MAASV_VOYAGE_API_KEY	If using Voyage	Embedding generation
MAASV_API_KEY	Optional	Protects maasv server endpoints with auth

For fully local operation (no cloud calls), use ollama as your embed provider and a local LLM. maasv is optimized for Qwen3-Embedding-8B via Ollama, with built-in Matryoshka dimension truncation. See the maasv README for local setup.

Data & Network Behavior

• maasv has no cloud service. The server runs on your machine, the database is a SQLite file on your disk. You own all of it.
• The only external calls are to your own LLM/embedding provider (Anthropic, OpenAI, Voyage) — using your own API keys, from your own machine. If you use ollama, zero data leaves your machine.
• The plugin talks only to localhost (127.0.0.1:18790). It makes no external network calls.
• autoCapture sends conversation summaries to your local maasv server for entity extraction. Extracted entities are stored in your local SQLite database.
• autoRecall reads from your local SQLite database and injects relevant memories into the agent's context.
• No telemetry, no analytics, no phone-home. maasv does not collect or transmit any data.

What You Get

• memory_search — 3-signal retrieval across your memory store
• memory_store — Dedup-aware memory storage
• memory_forget — Permanent deletion
• memory_graph — Knowledge graph: entity search, profiles, relationships
• memory_wisdom — Log reasoning, record outcomes, search past decisions

Links

• Plugin (npm): @maasv/openclaw-memory
• Server + core (PyPI): maasv
• Source: github.com/ascottbell/maasv