Hybrid Memory

Hybrid memory strategy combining OpenClaw's built-in vector memory with Graphiti temporal knowledge graph. Use when you need to recall past context, answer temporal questions ("when did X happen?"), or search memory files. Provides decision framework for when to use memory_search vs Graphiti.

Security Scan

The skill is an instruction-only guide describing when to use OpenClaw's memory_search vs an external Graphiti temporal store; its requirements and instructions are coherent with that purpose, but it assumes external scripts/infrastructure that you should verify before use.

Purpose & Capability

The name and description (hybrid memory combining vector memory and Graphiti) match the runtime instructions. The SKILL.md focuses on selecting between memory_search and Graphiti and gives call examples; nothing requested or instructed is unrelated to the stated purpose.

Instruction Scope

Instructions direct the agent to run memory_search and external scripts (graphiti-search.sh, graphiti-log.sh) and to deploy Graphiti via Docker. That is appropriate for the stated hybrid-memory purpose, but the skill assumes those scripts and the Graphiti service exist and that an embedding provider is configured. The instructions also advise editing AGENTS.md. There are no instructions to read unrelated system files or exfiltrate arbitrary data.

Install Mechanism

No install spec and no code files — instruction-only skill. This minimizes disk-write risk. The README points to a GitHub repo for setup, but the skill itself does not download or install anything automatically.

Credentials

The SKILL.md recommends configuring an embedding provider (mentions 'Gemini recommended') and deploying Graphiti, which in practice will require API keys/credentials and infrastructure, but the skill does not declare any required env vars. This omission is not necessarily malicious, but you should expect that following the guidance will require credentials for the embedding provider and any Graphiti endpoints.

Persistence & Privilege

Skill is not always-on and is user-invocable; it does not request persistent elevated privileges nor attempt to modify other skills. Advising edits to AGENTS.md is a normal operational step for configuring agent behavior.

Assessment

This skill is a usage guide, not executable code, and is coherent for its stated purpose — but it assumes you have or will deploy external components (Graphiti Docker stack, sync daemons, and shell scripts like graphiti-search.sh/log.sh) and an embedding provider (the doc mentions Gemini). Before installing/using: 1) Verify the referenced scripts (graphiti-*.sh) exist in your agent environment or replace them with safe equivalents; 2) Inspect the GitHub repo and any scripts you will run from it before executing or deploying Docker services; 3) Prepare and secure any embedding-provider/API credentials (store them in your secret manager, not in plain AGENTS.md); 4) Test in a sandboxed environment first since the setup will involve network services and Docker; 5) If you need the skill to run autonomously, ensure the external Graphiti endpoints and scripts are trustworthy — the skill itself provides only guidance and will rely on whatever external code you deploy.

Hybrid Memory System

Two memory systems, each with different strengths. Use both.

When to Use Which

Question Type	Tool	Example
Document content	memory_search	"What's in GOALS.md?"
Curated notes	memory_search	"What are our project guidelines?"
Temporal facts	Graphiti	"When did we set up Slack?"
Conversations	Graphiti	"What did the user say last Tuesday?"
Entity tracking	Graphiti	"What projects involve Alice?"

Quick Reference

memory_search (Built-in)

Semantic search over markdown files (MEMORY.md, memory/**/*.md).

memory_search query="your question"

Then use memory_get to read specific lines if needed.

Graphiti (Temporal)

Search for facts with time awareness:

graphiti-search.sh "your question" GROUP_ID 10

Log important facts:

graphiti-log.sh GROUP_ID user "Name" "Fact to remember"

Common group IDs:

• main-agent — Primary agent
• user-personal — User's personal context

Recall Pattern

When answering questions about past context:

1. Temporal questions → Check Graphiti first
2. Document questions → Use memory_search
3. Uncertain → Try both, combine results
4. Low confidence → Say you checked but aren't sure

AGENTS.md Template

Add to your AGENTS.md:

### Memory Recall (Hybrid)

**Temporal questions** ("when?", "what changed?", "last Tuesday"):
```bash
graphiti-search.sh "query" main-agent 10

Document questions ("what's in X?", "find notes about Y"):

memory_search query="your query"

When answering past context: check Graphiti for temporal, memory_search for docs.

## Setup

Full setup guide: https://github.com/clawdbrunner/openclaw-graphiti-memory

**Part 1: OpenClaw Memory** — Configure embedding provider (Gemini recommended)
**Part 2: Graphiti** — Deploy Docker stack, install sync daemons