Skip to content
Muzzy5150Public

About

Autonomous AI counterintelligence — MCP security, behavioral monitoring, memory integrity, threat hunting and forensic evidence.

Topics

Resources

Security policy

Stars

0 stars

Watchers

0 watching

Forks

Latest commit

 

History

8 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Official Ghost Ops pixel ghost mascot

GHOST OPS

Detect the rogue. Trace the behavior. Protect the memory.

Autonomous AI counterintelligence. Verify tool requests, investigate suspicious behavior, protect persistent memory and preserve an inspectable evidence trail.

GitHub repository Explore the Ghost Ops website

TypeScript Next.js React Model Context Protocol

Guild.ai verified execution ClickHouse verified execution Akash verified execution

Open Workstation · Capabilities · Architecture · Sponsor Evidence · Quick Start · Documentation

Ghost Ops public Operations workspace: agent network, recorded timeline and contextual inspector

Operations: an interactive agent network, movable windows and an inspectable security timeline.

Overview

Ghost Ops checks authenticated agent tool requests before supported operations execute. Its local MCP gateway enforces identity, session, credential, tool, resource and containment boundaries. SQLite preserves decisions and evidence; the workstation connects behavioral signals, memory integrity and investigations.

Note

The public workstation uses clearly labeled recorded and simulated data. Actual enforcement runs locally. Public provisioning, quarantine, restoration, scans, inference and publishing are disabled.

Core Capabilities

Capability What it does
AgentDNA Compares authenticated activity with behavioral baselines and explains deviations.
ShadowWatch Verifies identity and permissions; records rogue-agent indicators.
MemoryGuard Signs memory content and provenance; inspects versions and verified restoration.
GhostTrap Uses inert synthetic decoys to gather evidence of interest.
Ghost Hunt & Response Correlates cross-agent evidence and enforces scoped containment.
MCP gateway Authorizes bounded handlers before effects; denied requests do not execute.
Web Sentinel Retrieves approved security intelligence and preserves source references.
Forensic evidence Checks export integrity and authentication under explicit local-key trust.

How It Works

flowchart LR
  A[AI Agent] --> G[Authenticated MCP Gateway]
  G --> P{Policy Decision}
  P -->|Allowed| T[Bounded Tool Handler]
  P -->|Denied| D[No Handler Execution]
  T --> E[SQLite Security Events]
  D --> E
  E --> H[Ghost Hunt Investigation]
  H --> R[Containment & Forensic Evidence]
  E -. Sanitized telemetry .-> C[ClickHouse Analytics]
  classDef green fill:#102717,stroke:#79FF50,color:#ECF2E8;
  classDef denied fill:#261515,stroke:#E57979,color:#ECF2E8;
  class A,G,P,T,E,H,R,C green;
  class D denied;
Loading

The gateway and SQLite are authoritative. Sponsor intelligence, analytics and public digest checks support investigations; they do not grant permissions.

Inside the Workstation

The system

Ghost Ops homepage with the official animated pixel ghost

Investigation and memory integrity

Ghost Ops Incident Room: recorded synthetic case, agent relationships and illustrative memory integrity

Public sample case: recorded chronology and illustrative memory state, with privileged actions disabled.

Verified Sponsor Executions

Completed executions from the shared local investigation on 2026-10-09:

Technology Actual result Scope
Guild.ai The Smith completed investigator session 01a122d9-1384-351a-0000-5f4cefefda20; its policy read was authorized. Operator-mediated, bounded local bridge.
ClickHouse Cloud insert and query reconciled 47 event IDs to SQLite; 30 aggregate groups, server 26.6.1.2326. Sanitized telemetry; analytics do not decide authorization.
Akash Deployment 1791585638189 verified the shared receipt-summary digest and rejected an altered summary. Digest consistency, not private signer authenticity.

Recorded genuine Guild.ai, ClickHouse and Akash execution receipts

Badges describe verified completed executions, not current live connections. Submission evidence and limitations.

Technology

TypeScript, Next.js 16, React 19, Tailwind CSS, React Flow and Dagre. The local platform uses Prisma/SQLite, Zod and the official MCP SDK. Vercel hosts apps/website, reusing workstation presentation components. SDK and MCP proxy packages live in packages/.

Quick Start

Use Node.js 24 and npm in a fresh clone:

git clone https://github.com/Muzzy5150/ghost-ops.git
cd ghost-ops
npm ci
npm run setup
npm run sdk:build
npm run dev -- --port 3210

Open http://127.0.0.1:3210. Setup initializes demo data; use isolated state rather than an existing investigation database. Offline operation needs no paid service or model credentials. Use the provided server wrapper for administrative transport authority.

Security Scope

Enforcement applies to supported gateway-routed operations, not arbitrary OS activity. Behavioral and honeypot signals are evidence to investigate, not proof of malicious intent. Host or signing-key compromise defeats local evidence authentication. External transfers and model inference require explicit opt-in.

Documentation

About

Autonomous AI counterintelligence — MCP security, behavioral monitoring, memory integrity, threat hunting and forensic evidence.

Topics

Resources

Security policy

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages