Complete memory system + security layer for AI agents. Native OpenClaw integration. Persistent storage, semantic search, prompt injection firewall, credential protection, and audit trail. One package, full solution.
npm install shieldcortex
ShieldCortex gives your AI agent persistent memory AND protects it from attack. One package. Full solution.
Most AI agents are stateless — they forget everything between sessions. ShieldCortex fixes that with production-grade persistent memory. But memory creates risk: researchers have demonstrated memory poisoning attacks that hijack AI behaviour. ShieldCortex is the only solution that solves both problems.
``
┌─────────────────────────────────────────────────────────────┐
│ ShieldCortex │
├─────────────────────────────┬───────────────────────────────┤
│ MEMORY SYSTEM │ SECURITY LAYER │
├─────────────────────────────┼───────────────────────────────┤
│ • Persistent storage │ • Memory firewall │
│ • Semantic search │ • Prompt injection detection │
│ • Project scoping │ • Credential protection │
│ • Importance ranking │ • Sub-agent access control │
│ • Auto-cleanup & decay │ • Full audit trail │
│ • Category organisation │ • Threat quarantine │
└─────────────────────────────┴───────────────────────────────┘
Stop choosing between memory and security. Get both.
See every threat across every project in one team dashboard. The local package now syncs audit data to ShieldCortex Cloud — sign up directly from the built-in dashboard, no CLI commands needed.
`
Local Agent ShieldCortex Cloud
┌──────────────┐ ┌──────────────────────┐
│ npm package │──audit sync──▶│ Team dashboard │
│ (free, │ │ Audit log + stats │
│ unlimited) │ │ Team invites │
│ │ │ Usage analytics │
└──────────────┘ └──────────────────────┘
---
AI agents are configured by instruction files (SKILL.md, CLAUDE.md, .cursorrules, etc.) — and attackers are hiding prompt injections inside them. ShieldCortex scans all of them:
| Format | File |
|--------|------|
| Claude Code Skills | SKILL.md |CLAUDE.md
| Claude Code Config | |HOOK.md
| OpenClaw Hooks | , handler.js |.cursorrules
| Cursor | |.windsurfrules
| Windsurf | |.clinerules
| Cline | |copilot-instructions.md
| GitHub Copilot | |.aider.conf.yml
| Aider | |.continue/config.json
| Continue | |
`bashScan all instruction files on your machine
npx shieldcortex scan-skills
The dashboard Skills tab shows results with severity badges, expandable threat details, and actions:
- Trust — Mark known-safe skills so they're not flagged on future scans (free, local)
- Remove — Delete dangerous skill files from disk (cloud-connected, premium)
$3
- Direct injection — hidden in content
- Credential harvesting — Attempts to exfiltrate secrets
- Encoding tricks — Base64/hex/unicode payloads
- Slow-burn assembly — Attack fragments planted over days
- Privilege escalation — System command references
- Skill file poisoning — Hidden instructions in agent configuration files$3
Running multiple agents or sub-agents? ShieldCortex prevents rogue agents from accessing sensitive data:
| Feature | What It Does |
|---------|--------------|
| Hierarchical Trust Decay | Sub-agents get lower trust:
user-spawned (0.9) → >task-1 (0.63) → >subtask-2 (0.44) |
| Origin-Based Scoring | Different trust by source: user (0.9), cron (0.5), agent-spawned (0.3), web (0.2) |
| Credential Isolation | RESTRICTED memories blocked below trust 0.7 — sub-agents can't access secrets |
| Depth Circuit Breaker | Agents beyond depth 5 get trust = 0 (blocked entirely) |
| Auto-Quarantine | Low-trust writes go to quarantine for human review |
| Environment Detection | Auto-detects sub-agents from CLAUDE_CODE_ENTRYPOINT — zero config |`
Trust ≥0.7 → Read all, write direct, delete own
Trust 0.5–0.7 → Read own + non-restricted, quarantine writes
Trust <0.5 → Read own only, quarantine only, no delete
Result: A sub-agent spawning another sub-agent that tries to read your API keys? Blocked.
---
Quick Start
bash
Install globally
npm install -g shieldcortexAuto-configure for your agent (Claude Code, OpenClaw, LangChain, etc.)
npx shieldcortex setupThat's it. Your agent now has persistent memory + security.
`Already using Claude Cortex? Migrate in one command:
bash
npx shieldcortex migrate
`Verify installation:
bash
npx shieldcortex doctor
`---
OpenClaw Setup
Three commands. That's it. Works on macOS, Linux (including headless servers), and Windows.
bash
1. Install globally
npm install -g shieldcortex2. Install the OpenClaw hook
sudo npx shieldcortex openclaw install3. Connect to Cloud (get your API key from the Cloud dashboard)
npx shieldcortex config --cloud-api-key --cloud-enable
`Every memory your OpenClaw agent saves now passes through the 5-layer defence pipeline, and audit data syncs to your team's Cloud dashboard automatically.
$3
Adding someone to your team? Give them an API key and these three commands:
bash
npm install -g shieldcortex
sudo npx shieldcortex openclaw install
npx shieldcortex config --cloud-api-key --cloud-enable
`No signup, no browser, no account creation needed. The API key ties their scans to your team. Their device appears in your Cloud dashboard automatically once scans start flowing.
> Headless servers? No problem — ShieldCortex runs entirely via CLI. No GUI or browser required.
---
Supported Agents
ShieldCortex is agent-agnostic middleware:
| Agent | Integration |
|-------|-------------|
| Claude Code |
— Native MCP server |
| OpenClaw / Moltbot | npx shieldcortex openclaw install — Native hooks |
| LangChain JS | import { ShieldCortexMemory } from 'shieldcortex/integrations/langchain' |
| Python (CrewAI, AutoGPT) | REST API — POST /api/v1/scan |
| Any MCP agent | Via MCP protocol or @langchain/mcp-adapters |If your agent stores memories, ShieldCortex can power and protect them.
---
Requirements
| Requirement | Version |
|-------------|---------|
| Node.js | >= 18.0.0 |
| Platform | macOS, Linux, or Windows |
| Storage | ~50MB for SQLite database |
No external dependencies. No Docker required. Just
npm install and go.---
Is Your AI Agent Already Compromised?
Find out in 30 seconds:
`bash
npx shieldcortex setup
`Then ask your agent: "Scan my memories for threats"
ShieldCortex will scan every stored memory and report:
- Hidden instructions disguised as normal content
- Credential harvesting attempts
- Encoded payloads (base64, unicode, hex)
- Fragmented attack patterns spread across memories
- Privilege escalation attempts
No threats found? Great — now you're protected going forward.
---
LangChain Integration
javascript
import { ShieldCortexMemory, ShieldCortexGuard } from 'shieldcortex/integrations/langchain';// As a LangChain memory backend (scans before storing)
const memory = new ShieldCortexMemory({ mode: 'balanced' });
const vars = await memory.loadMemoryVariables({ input: 'deployment config' });
await memory.saveContext({ input: 'hello' }, { output: 'hi' });
// As standalone middleware (scan without storing)
const guard = new ShieldCortexGuard();
const result = guard.scan('some content to check');
if (!result.allowed) {
console.warn('Blocked:', result.firewall.reason);
}
`---
REST API (Any Agent)
Start the API server, then scan content from any language or framework:
bash
npm run dev:api # Starts on http://localhost:3001
`bash
Scan content
curl -X POST http://localhost:3001/api/v1/scan \
-H 'Content-Type: application/json' \
-d '{"content": "memory to scan", "title": "test"}'Batch scan
curl -X POST http://localhost:3001/api/v1/scan/batch \
-H 'Content-Type: application/json' \
-d '{"items": [{"content": "item 1"}, {"content": "item 2"}]}'Query audit logs
curl http://localhost:3001/api/v1/audit?firewallResult=BLOCKList quarantined items
curl http://localhost:3001/api/v1/quarantineApprove/reject quarantined items
curl -X POST http://localhost:3001/api/v1/quarantine/1/approve
curl -X POST http://localhost:3001/api/v1/quarantine/1/reject
`---
OpenClaw Integration
One command. Persistent memory for every OpenClaw session.
bash
sudo npx shieldcortex openclaw install
`This installs the cortex-memory hook directly into OpenClaw's bundled hooks directory. No configuration needed.
$3
Session Start During Session Session End (/new)
│ │ │
▼ ▼ ▼
┌─────────────┐ ┌─────────────┐ ┌─────────────┐
│ Inject past │ │ "remember │ │ Auto-extract│
│ context │ │ this: ..." │ │ decisions, │
│ into agent │ │ → save │ │ fixes, etc. │
└─────────────┘ └─────────────┘ └─────────────┘
| Feature | What Happens |
|---------|--------------|
| Context Injection | On session start, relevant past memories are injected into the agent's bootstrap context |
| Keyword Triggers | Say "remember this:" or "don't forget:" followed by content to save it with critical importance |
| Auto-Extraction | On
, the hook extracts architecture decisions, bug fixes, and learnings from the ending session |
| Security | All content passes through the 5-layer defence pipeline before storage |$3
The database (
~/.shieldcortex/memories.db) is shared with Claude Code. Memories created in OpenClaw appear instantly in Claude Code sessions, and vice versa.$3
`bash
sudo npx shieldcortex openclaw install # Install hook
sudo npx shieldcortex openclaw uninstall # Remove hook
npx shieldcortex openclaw status # Check status
`---
Dashboard
bash
npx shieldcortex --dashboard
`- Dashboard: http://localhost:3030
- API: http://localhost:3001
Views: Shield (defence overview), Audit Log, Quarantine, Memories, 3D Brain, Knowledge Graph, Skills Scanner.
$3
Connect your local dashboard to ShieldCortex Cloud and see threats from all your projects in one place. The dashboard includes a guided setup — just enter your email and click the magic link.
bash
Or configure via CLI
npx shieldcortex config --cloud-api-key --cloud-enable
`Once connected, every local scan automatically syncs audit metadata (not content) to your Cloud dashboard.
$3
bash
npx shieldcortex service install # Enable
npx shieldcortex service uninstall # Disable
npx shieldcortex service status # Check
`Works on macOS (launchd), Linux (systemd), and Windows.
---
CLI Reference
bash
npx shieldcortex setup # Auto-detect agent + configure hooks
npx shieldcortex install # Alias for setup
npx shieldcortex migrate # Migrate from Claude Cortex
npx shieldcortex doctor # Check installation health
npx shieldcortex scan-skills # Scan all agent instruction files
npx shieldcortex scan-skill # Scan a specific instruction file
npx shieldcortex --dashboard # Start dashboard + API
npx shieldcortex --version # Show version
npx shieldcortex service install # Auto-start on login
npx shieldcortex graph backfill # Extract entities from existing memories
npx shieldcortex openclaw install # Install OpenClaw hook
npx shieldcortex config --cloud-api-key # Set Cloud API key
npx shieldcortex config --cloud-enable # Enable cloud sync
npx shieldcortex uninstall # Full uninstall (requires confirmation)
npx shieldcortex uninstall --confirm # Non-interactive uninstall
`---
MCP Tools
| Tool | Description |
|------|-------------|
|
| Store a memory (optional — hooks do this automatically) |
| recall | Search memories by query, category, or tags |
| forget | Delete memories |
| get_context | Get relevant project context |
| memory_stats | View memory statistics |
| scan_memories | Scan existing memories for threats |
| audit_query | Query the defence audit trail |
| quarantine_review | Review quarantined memories |
| defence_stats | Threat counts, trust distribution |
| graph_query | Traverse the knowledge graph |
| graph_entities | List known entities |
| graph_explain` | Find paths between entities |---
| Feature | Claude Cortex | Mem0 | Zep | ShieldCortex |
|---------|---------------|------|-----|------------------|
| Persistent Memory | Yes | Yes | Yes | Yes |
| Semantic Search | Yes | Yes | Yes | Yes |
| Prompt Injection Detection | No | No | No | Yes |
| Memory Firewall | No | No | No | Yes |
| Sub-Agent Access Control | No | No | No | Yes |
| Audit Trail | No | No | Partial | Yes |
| Credential Protection | No | No | No | Yes |
| Cloud Team Dashboard | No | Yes | Yes | Yes |
| MCP Native | Yes | No | No | Yes |
| Self-Hosted | Yes | No | Partial | Yes |
| Open Source | Yes | Partial | Partial | Yes |
ShieldCortex is the only memory system built for adversarial conditions.
---
| Tier | What You Get | Price |
|------|--------------|-------|
| Free | npm package (unlimited local scans) + Cloud (500 scans/month, 1 member, 7-day audit retention) | Free |
| Pro | 10K cloud scans/month, team invites, 90-day retention | £29/mo |
| Team | 100K cloud scans/month, unlimited members, 1-year retention | £99/mo |
| Enterprise | Self-hosted, SLA, custom rules, dedicated support | Contact us |
The npm package is free and unlimited for local use. Cloud tiers add team visibility, longer retention, and higher scan limits. See shieldcortex.ai/pricing for details.
---
- Website: shieldcortex.ai
- Cloud Dashboard: api.shieldcortex.ai
- npm: npmjs.com/package/shieldcortex
- GitHub: github.com/Drakon-Systems-Ltd/ShieldCortex
- Docs: shieldcortex.ai/docs
- Architecture: ARCHITECTURE.md
---
MIT
Built by Drakon Systems