OpenClaw Plugin

MIND for OpenClaw

Persistent knowledge graph memory for AI agents. 11 tools, 4 skills, full lifecycle hooks. The most complete memory plugin available.

The MIND coral reef sign with the OpenClaw lobster mascot looking up at it from the ocean floor
Get a MIND API key → View on GitHub
openclaw plugins install @astramindapp/openclaw-mind
11
Tools
4
Skills
2
Hooks
50+
AI Models

Other memory plugins store text. MIND understands it.

MIND is the only OpenClaw memory plugin built on a true knowledge graph (LightRAG) with patented MINDsense emotional intelligence, life management, CRM, and graph-traversal queries that vector-only tools cannot match.

How MIND Compares

Capability MIND Mem0 Anthropic Memory MCP
Tools1188
Skills (agent protocols)420
True knowledge graph✓ LightRAG✗ vectors✗ JSON
50+ AI models✓✗✗ Claude only
Emotional intelligence (patented)✓ MINDsense✗✗
Graph traversal queries✓✗✗
Auto-recall before each turn✓✓✗
Auto-capture after each turn✓✓✗
Life management (tasks, goals)✓✗✗
CRM (contacts, activities)✓✗✗
Mobile app for the same data✓✗✗
Free tier✓✓✓

Quick Start

1

Install the plugin

openclaw plugins install @astramindapp/openclaw-mind

Requires Node.js 22 or higher and OpenClaw 2026.3.28+.

2

Get a MIND API key

Sign up at m-i-n-d.ai (free tier — no card required), then navigate to Settings → Developer → Create API Key. Your key starts with mind_.

3

Initialize and verify

openclaw mind init --api-key mind_your_key_here
openclaw mind status

You should see MIND: CONNECTED with your entity count, credits, and feature flags.

Auto-recall and auto-capture are now active. Every agent turn will automatically pull relevant memories from your MIND knowledge graph and store new facts after.

11 Tools

Once installed, your OpenClaw agent gains 11 native tools — 6 standard CRUD operations plus 5 unique to MIND.

Memory CRUD (parity)
mind_search
Hybrid semantic + graph search across the knowledge graph. Returns AI-synthesized answers grounded in stored memories.
mind_add
Store a memory. Auto-extracts entities + relationships + (optionally) MINDsense emotional weights.
mind_get
Retrieve a single memory by ID with full content, tags, and metadata.
mind_list
Paginated list of stored memories filtered by type, tag, or date.
mind_update
Update an existing memory. KG is re-extracted from new content automatically.
mind_delete
Soft delete. Removes the memory and the entities/relationships sourced from it.
MIND-Only Capabilities
mind_query_graph
Graph traversal queries — "who is connected to X", "what concepts cluster around Y". Vector-only memory tools cannot do this.
mind_recall_emotional
Emotionally weighted recall via MINDsense (patented). Surfaces memories that mattered to the user, not just textually similar ones.
mind_context
Loads structured persistent context: soul (identity), user (preferences), rules (constraints), priorities (current goals), recent (last 7 days). Call at session start.
mind_life
Create and manage tasks, goals, projects. Synced to the MIND web and mobile apps.
mind_crm_log
Log contacts and interaction history (calls, emails, meetings). Captured automatically from conversation context.

4 Agent Skills

Skills are agent protocols that ship with the plugin. They give your agent rules for when to capture, when to recall, and how to consolidate memory over time.

memory-triage
5-gate decision for capture: FUTURE UTILITY, NOVELTY, FACTUAL, SAFE, EMOTIONAL SALIENCE. Most turns produce zero memory operations — that's correct.
memory-dream
Periodic consolidation. Reviews memories, merges duplicates, removes noise, rewrites unclear entries. Like sleep consolidates short-term to long-term in biological brains.
mind-emotional-encoding
Assigns MINDsense valence/arousal weights to captured memories. High-emotion content gets deeper KG encoding (patented MIND-PAT-001).
mind-graph-recall
Graph-traversal recall protocol. Finds memories by walking entity relationships, not just text similarity.

Lifecycle Hooks

The plugin registers two hooks into OpenClaw's agent lifecycle. These run automatically — no agent prompting required.

↑

auto-recall (before each turn)

Hooks before_prompt_build. Runs a hybrid KG search using the latest user message as the query. Injects the synthesized result via prependContext so the model sees relevant memories before responding. Top-K and search threshold are configurable.

↓

auto-capture (after each turn)

Hooks agent_end. Filters noise messages, then sends the conversation to MIND for entity + relationship + MINDsense extraction. Gated by the memory-triage skill so most turns capture nothing — only durable, novel, factual, safe, emotionally relevant facts get stored.

Per-agent isolation: subagent sessions get ephemeral namespaces and skip both hooks. Memories are scoped by sessionKey so subagents don't pollute the main agent's KG.

CLI Commands

The plugin adds native openclaw mind subcommands.

# Initial setup
openclaw mind init --api-key mind_your_key_here

# Verify connection and show settings
openclaw mind status

# Migrate existing OpenClaw workspace memory to MIND
openclaw mind import

# Quick CLI search of the knowledge graph
openclaw mind search "what did we decide about pricing?"

Configuration

All settings are optional. Defaults are sensible. Configure via ~/.openclaw/plugins/openclaw-mind.yaml or env vars.

apiKey: ${MIND_API_KEY}            # required
baseUrl: https://www.m-i-n-d.ai  # override only for self-hosting
autoCapture: true               # extract facts after each turn
autoRecall: true                # inject memories before each turn
topK: 10                       # memories to inject per turn
searchThreshold: 0.5            # similarity threshold (0-1)
queryMode: hybrid               # hybrid | mix | global | local | naive
enableMindsense: true           # MINDsense emotional weighting
enableLifeIntegration: true     # allow agent to create life items
enableCrmLogging: true          # allow agent to log contacts

Patents Pending

MIND's underlying technology is protected by multiple provisional patent applications:

Also Available: Universal MCP Server

If you're using Claude Desktop, Cursor, Windsurf, VS Code, or any other MCP-compatible AI tool, we also ship the MIND MCP Server — same backend, MCP protocol instead of OpenClaw plugin. Install both: they share the same MIND knowledge graph so memories captured in one show up in the other.

Links & Resources

Get Started

Related