Tools
Memex Openclaw
MeMex Zero RAG ContextEngine plugin for OpenClaw โ inject compiled wiki knowledge into every agent session
Install
npm install
pnpm
Configuration Example
{
"plugins": {
"slots": {
"contextEngine": "memex-openclaw"
}
}
}
README
# @memex/openclaw-memex โ MeMex Zero RAG ContextEngine Plugin
> Durable, anti-hallucination wiki memory for OpenClaw agents. Injects compiled knowledge from MeMex Zero RAG's wiki into every assembly pass.
## What It Does
OpenClaw agents lose context between sessions like everyone else. This plugin solves that by connecting OpenClaw to **MeMex Zero RAG** โ a structured wiki knowledge system with built-in anti-hallucination, knowledge decay, and conflict detection.
Every time OpenClaw assembles context for a reply, the plugin:
1. **Searches the MeMex wiki** for relevant entries based on conversation context
2. **Injects matched knowledge** as a system prompt supplement
3. **Respects knowledge decay** โ expired entries automatically deprioritize
4. **Flags conflicts** โ duplicate or contradictory entries are reported on bootstrap
The result: your agent never forgets documented decisions, and never hallucinates institutional knowledge it should have known.
## Requirements
| Requirement | Version |
|---|---|
| OpenClaw Gateway | โฅ 2026.5.28 |
| Node.js | โฅ 22 |
## Installation
```bash
openclaw plugins install @jpeetz/openclaw-memex
```
Or from a local build:
```bash
cd /path/to/memex-openclaw-plugin
pnpm build
openclaw plugins install --link /path/to/memex-openclaw-plugin
```
Restart the gateway:
```bash
openclaw gateway restart
```
## Configuration
The plugin is configured via `plugins.entries.memex-openclaw` in your `openclaw.json`:
```json5
{
"plugins": {
"slots": {
"contextEngine": "memex-openclaw"
},
"entries": {
"memex-openclaw": {
"enabled": true,
"config": {
// Path to MeMex Zero RAG wiki
"wikiPath": "~/workspace/MeMex-Zero-RAG/wiki/",
// Max wiki entries to inject per assembly (default: 5)
"maxInjectEntries": 5,
// Max chars per injected entry (default: 2000)
"maxEntryChars": 2000,
// Search mode: "keyword" | "semantic" | "hybrid" (default: "keyword")
"searchMode": "keyword",
// Whether to respect knowledge decay dates (default: true)
"respectDecayDates": true,
// Days before decay penalty kicks in (default: 90)
"decayThresholdDays": 90,
// Wiki sections to prioritize (default: ["concepts/", "agentforge/"])
"priorityPaths": ["concepts/", "agentforge/"],
// Enable conflict detection on bootstrap (default: true)
"enableConflictDetection": true,
// Verbose logging (default: false)
"verbose": false
}
}
}
}
}
```
### Slot Assignment
The plugin registers as a ContextEngine provider. Set the slot in your config:
```json
{
"plugins": {
"slots": {
"contextEngine": "memex-openclaw"
}
}
}
```
If you want to use MeMex alongside another ContextEngine plugin, disable MeMex's `injectOnAssemble` and run it purely as a conflict detector:
```json
{
"config": {
"injectOnAssemble": false,
"enableConflictDetection": true
}
}
```
## How It Works
### Architecture
```
User Message
โ
โผ
ContextEngine.assemble โโโบ WikiIndex.search(query)
โ โ
โ โโโโโโโโดโโโโโโโ
โ โผ โผ
โ BM25 Search Priority Boost
โ โ โ
โ โโโโโโโโฌโโโโโโโ
โ โผ
โ Top-K Results
โ โ
โผ โผ
System Prompt Supplement โโโบ Main Agent Reply
```
### Hook Lifecycle
| Hook | What MeMex Does |
|---|---|
| `bootstrap` | Loads config, builds wiki index, runs conflict detection |
| `ingest` | Scans messages for new knowledge indicators (passive) |
| `assemble` | Searches wiki, injects relevant entries as system prompt supplement |
| `compact` | Adjusts session knowledge window based on decay awareness |
| `afterTurn` | Logs injection statistics per session |
| `prepareSubagent` | Provides lightweight wiki context for subagent tasks |
| `onSubagentEnded` | (Reserved for future use) |
### Search Algorithm
The plugin uses **BM25 keyword search** with:
- **Title boost**: Matches in entry titles score 2ร higher
- **Priority paths**: Entries under `concepts/` and `agentforge/` receive a +100 score boost
- **Decay penalty**: Expired entries score at 30%; entries within 30 days of decay score at 50%โ100%
- **Document frequency**: Rare terms score higher (standard IDF)
## Bundled Skill
The plugin installs a `memex-search` skill that agents can use for explicit wiki queries. Agents naturally learn to use it through the injected context.
## Verification
Check plugin status:
```bash
openclaw plugins list
openclaw plugins inspect memex-openclaw
```
The agent's `/context list` will show injected wiki entries as:
```
MeMex Zero RAG Wiki Context: OK | injected 5 entries (~850 chars)
```
## Development
```bash
git clone https://github.com/JPeetz/memex-openclaw-plugin
cd memex-openclaw-plugin
pnpm install
pnpm build
pnpm test
```
For local testing with OpenClaw:
```bash
openclaw plugins install --link /path/to/memex-openclaw-plugin
```
## Relation to Hermes Agent
This plugin uses the same core MeMex Zero RAG engine as the Hermes Agent plugin at `@memex/hermes-memory-provider` but adapted for OpenClaw's TypeScript/ContextEngine architecture. Knowledge indexed in one platform is immediately available in the other.
## License
Apache 2.0
tools
Comments
Sign in to leave a comment