OmniGraph

Know what you've shared with AI. OmniGraph is an open-source AI interaction tracker that gives you a timeline and knowledge graph of every message you send to ChatGPT, Claude, Gemini, Copilot, and other AI tools.

A browser extension silently captures outbound messages. A dashboard classifies them by sensitivity, maps data subjects, and visualizes your cumulative AI exposure — so you always know what went where.

Why

AI tools are everywhere. You paste code into ChatGPT, share financials with Claude, drop resumes into Copilot. But nobody tracks the cumulative picture:

What sensitive data have you shared?
With which services?
When did it happen?
How exposed are specific people, projects, or credentials?

OmniGraph answers these questions with a passive browser extension and a visual dashboard. No proxies, no browser profile changes — just install and browse.

How It Works

┌─────────────────────────────────────────────────────────────────┐
│                   BROWSER EXTENSION (MV3)                       │
│                                                                 │
│  ┌─────────────┐   ┌──────────┐   ┌──────────────────────────┐ │
│  │ Interceptor  │   │  Bridge  │   │   Background Worker      │ │
│  │ (MAIN world) │──▶│(ISOLATED)│──▶│ Dedup · Batch · Retry    │ │
│  │ fetch / XHR  │   │  relay   │   │ IndexedDB persistence    │ │
│  │ WebSocket    │   │          │   │                          │ │
│  └─────────────┘   └──────────┘   └───────────┬──────────────┘ │
│  ┌─────────────┐                               │                │
│  │ DOM Observer │───────────────────────────────┘                │
│  │ Input / Paste│                                               │
│  └─────────────┘                                                │
└─────────────────────────────────────────────────────────────────┘
                                │
                          POST /api/ingest
                                │
┌─────────────────────────────────────────────────────────────────┐
│                     NEXT.JS BACKEND                             │
│                                                                 │
│  ┌─────────────────┐   ┌──────────────┐   ┌──────────────────┐ │
│  │ Ingest API       │   │  Classifier  │   │  Graph API       │ │
│  │ Zod validation   │──▶│  OpenAI      │──▶│  CRUD + Search   │ │
│  │ Rate limiting    │   │  Queue       │   │  Timeline        │ │
│  │ Bearer auth      │   │  Retry       │   │  Stats           │ │
│  └─────────────────┘   └──────────────┘   └──────────────────┘ │
│                                                                 │
│  ┌──────────────────────────────────────────────────────────┐   │
│  │                     FalkorDB                              │   │
│  │  Service ← SENT_TO ← Disclosure → MENTIONS → DataSubject │   │
│  │                         │                                 │   │
│  │                    CONTAINS → DataCategory                │   │
│  └──────────────────────────────────────────────────────────┘   │
└─────────────────────────────────────────────────────────────────┘
                                │
                          Dashboard UI
                                │
┌─────────────────────────────────────────────────────────────────┐
│                        DASHBOARD                                │
│                                                                 │
│  Dashboard         Timeline         Exposure Graph              │
│  Stats & trends    Chronological    Interactive knowledge graph  │
│  Service breakdown Search & filter  Node inspector              │
│  Sensitivity bars  CSV export       Zoom & pan                  │
└─────────────────────────────────────────────────────────────────┘

Detection Architecture

OmniGraph uses dual-layer capture for reliability:

Network interception — Monkey-patches fetch, XMLHttpRequest, and WebSocket in the page's MAIN world to capture outbound API calls to AI services.
DOM observation — A MutationObserver watches for editable elements (textareas, contenteditable divs) and captures input/paste events as a fallback.

Events from both layers are deduplicated by content hash within a 3-second window, batched, persisted to IndexedDB, and sent to the backend with exponential-backoff retry.

Supported AI Services

Service	Network Capture	DOM Capture
ChatGPT (chat.openai.com, chatgpt.com)	✅	✅
Claude (claude.ai)	✅	✅
Gemini (gemini.google.com)	✅	✅
Copilot (copilot.microsoft.com)	✅	✅
Perplexity (perplexity.ai)	✅	✅
Harvey (app.harvey.ai)	✅	✅
Clay (app.clay.com)	✅	✅
Jasper (app.jasper.ai)	✅	✅
Notion AI (notion.so)	✅	✅

Adding a new service requires a small entry in extension/src/services/registry.js.

Quick Start

Prerequisites

Tool	Version	Check
Node.js	20+	`node --version`
Docker	24+	`docker --version`
Chrome/Chromium	Latest	—
OpenAI API Key	—	platform.openai.com

1. Clone and install

git clone https://github.com/ASR4/omni-graph.git
cd omni-graph
npm install

2. Start FalkorDB

docker compose up -d

Verify:

docker exec omnigraph-falkordb redis-cli -a omnigraph-dev-password ping
# Expected: PONG

3. Configure environment

cp .env.example .env

Set your OpenAI API key in .env:

OPENAI_API_KEY=sk-proj-your-key-here

Everything else has working defaults. See Environment Variables for the full reference.

4. Start the dev server

npm run dev

Dashboard is at http://localhost:3001.

5. Load the browser extension

Open chrome://extensions in Chrome
Enable Developer mode (top-right toggle)
Click Load unpacked
Select the extension/ directory from this repo

The OmniGraph icon appears in your toolbar. Click it to see the popup with status and controls.

6. Use any AI tool

Open ChatGPT, Claude, or any supported service and send a message. Within seconds, it appears on your OmniGraph dashboard timeline.

Environment Variables

Variable	Purpose	Required	Default
`OPENAI_API_KEY`	Classification via OpenAI	Yes	—
`INGEST_API_KEY`	Shared secret for extension auth	No	(auth disabled)
`FALKORDB_HOST`	Graph database host	No	`localhost`
`FALKORDB_PORT`	Graph database port	No	`6379`
`FALKORDB_PASSWORD`	FalkorDB/Redis password	No	`omnigraph-dev-password`
`FALKORDB_GRAPH_NAME`	Graph name in FalkorDB	No	`omnigraph`
`AUTH_SECRET`	NextAuth JWT signing secret	Yes (prod)	`change-me-to-a-random-secret`
`OMNIGRAPH_PASSWORD`	Dashboard login password	No	(auth disabled)
`LOG_LEVEL`	Pino log level	No	`info`

Generate secrets with: openssl rand -base64 32

If INGEST_API_KEY is set, the extension must send it as a Bearer token. If unset, the ingest endpoint accepts all requests (suitable for local dev).

Project Structure

omni-graph/
├── extension/                          # Chrome MV3 browser extension
│   ├── manifest.json                   # Extension config and permissions
│   ├── icons/                          # Extension icons (16, 48, 128px)
│   └── src/
│       ├── interceptor.js              # MAIN world: fetch/XHR/WS monkey-patching
│       ├── bridge.js                   # ISOLATED world: relay to service worker
│       ├── background.js               # Service worker: dedup, batch, retry, IndexedDB
│       ├── popup.html                  # Toolbar popup UI
│       ├── popup.js                    # Popup logic and state
│       ├── services/
│       │   └── registry.js             # AI service definitions and body parsers
│       └── content/
│           ├── element-discovery.js    # DOM observer for input detection
│           └── adapters/              # Service-specific DOM adapters
│               ├── chatgpt.js
│               ├── claude.js
│               └── gemini.js
├── src/
│   ├── app/                            # Next.js App Router
│   │   ├── page.tsx                    # Dashboard: stats, breakdowns, recent activity
│   │   ├── timeline/page.tsx           # Timeline: chronological feed, search, export
│   │   ├── graph/page.tsx              # Graph: interactive exposure map
│   │   ├── login/page.tsx              # Authentication
│   │   └── api/
│   │       ├── ingest/route.ts         # Extension → server ingestion
│   │       ├── graph/route.ts          # Graph data, search, timeline, stats
│   │       ├── health/route.ts         # System health check
│   │       └── auth/[...nextauth]/     # NextAuth handlers
│   ├── components/
│   │   ├── graph/
│   │   │   ├── GraphCanvas.tsx         # Graph visualization container
│   │   │   ├── CosmographWrapper.tsx   # Cosmograph WebGL renderer
│   │   │   └── NodeDetail.tsx          # Node inspector panel
│   │   └── ui/                         # shadcn/ui primitives
│   ├── lib/
│   │   ├── types.ts                    # TypeScript type definitions
│   │   ├── falkordb.ts                 # FalkorDB client and Cypher helpers
│   │   ├── classifier.ts              # OpenAI-based content classification
│   │   ├── classifier-queue.ts        # Async classification queue with retry
│   │   ├── rate-limit.ts              # Token bucket rate limiter
│   │   ├── auth.ts                     # NextAuth configuration
│   │   ├── logger.ts                   # Pino structured logging
│   │   ├── api-error.ts               # Safe error responses
│   │   ├── constants.tsx              # UI constants (icons, colors)
│   │   └── graph/
│   │       ├── schema.ts              # Node/edge types, colors, sizes
│   │       └── operations.ts          # Graph CRUD, search, timeline, stats
│   ├── middleware.ts                   # Route protection
│   ├── instrumentation.ts             # Startup: re-queue unclassified disclosures
│   └── __tests__/                      # Vitest test suite
│       ├── rate-limit.test.ts
│       ├── classifier.test.ts
│       └── ingest-validation.test.ts
├── docker-compose.yml                  # FalkorDB with persistence and auth
├── Dockerfile                          # Multi-stage production build
├── vitest.config.ts
├── .env.example                        # Environment variable reference
└── package.json

Graph Data Model

OmniGraph stores disclosures in a property graph with four node types and four edge types:

(Service) ←[SENT_TO]← (Disclosure) →[MENTIONS]→ (DataSubject)
                            │
                       [CONTAINS]
                            ↓
                      (DataCategory)

(DataSubject) ←[EXPOSED_TO]→ (Service)

Node	Description	Example
Service	An AI tool or SaaS product	ChatGPT, Claude, Gemini
Disclosure	A single message sent to a service	"Here's our Q4 revenue..."
DataSubject	A person, company, project, or product mentioned	"John Smith", "Acme Corp"
DataCategory	Type of sensitive data contained	PII, Financial, Code, Credentials

Sensitivity levels: LOW, MEDIUM, HIGH, CRITICAL — assigned by the OpenAI classifier.

API Reference

`POST /api/ingest`

Receives disclosure batches from the browser extension. Validates with Zod, stores in FalkorDB, queues for classification.

{
  "disclosures": [
    {
      "id": "uuid",
      "service": "ChatGPT",
      "serviceUrl": "https://chatgpt.com",
      "endpointUrl": "https://chatgpt.com/backend-api/f/conversation",
      "messageText": "Here is our Q4 revenue breakdown...",
      "timestamp": 1710000000000,
      "contentType": "text",
      "inputMethod": "typed",
      "captureSource": "network",
      "metadata": {
        "messageLength": 42,
        "tabUrl": "https://chatgpt.com/c/abc123"
      }
    }
  ]
}

Headers: Authorization: Bearer <INGEST_API_KEY> (if configured)

`GET /api/graph?action=data&limit=500`

Returns graph nodes and edges for visualization.

`GET /api/graph?action=search&query=keyword&limit=20`

Search nodes by keyword.

`GET /api/graph?action=timeline&limit=50&offset=0`

Chronological disclosures with optional filters: service, sensitivity, search.

`GET /api/graph?action=dashboard`

Aggregate stats: totals, weekly count, service breakdown, sensitivity distribution, most-exposed subjects, recent activity.

`DELETE /api/graph?id=node-uuid`

Delete a node with cascade (removes connected edges and orphaned nodes).

`GET /api/health`

System health check (FalkorDB connectivity).

Testing

# Run the full suite
npm test

# Watch mode
npm run test:watch

# Full CI pipeline (typecheck + lint + test)
npm run ci

Test Coverage

File	Tests	Covers
`src/__tests__/rate-limit.test.ts`	8	Token bucket: burst, refill, per-client isolation, key expiry
`src/__tests__/classifier.test.ts`	6	Zod schema validation, category/sensitivity/subject parsing, edge cases
`src/__tests__/ingest-validation.test.ts`	6	Ingest payload validation, required fields, bounds checking

Build Commands

Command	What It Does
`npm run dev`	Start dev server with hot reload (port 3001)
`npm run build`	Next.js production build
`npm run start`	Start production server (port 3001)
`npm run typecheck`	TypeScript type checker
`npm run lint`	ESLint
`npm run test`	Run Vitest suite
`npm run test:watch`	Vitest in watch mode
`npm run ci`	Typecheck + lint + test

Tech Stack

Layer	Technology
Frontend	Next.js 15, React 19, Tailwind CSS 4, shadcn/ui
Graph Visualization	Cosmograph (WebGL)
Graph Database	FalkorDB (Cypher queries over Redis protocol)
Classification	OpenAI API with Zod-validated structured output
Auth	NextAuth v5
Validation	Zod
Logging	Pino
Testing	Vitest
Extension	Chrome Manifest V3

Common Gotchas

Port 6379 in use. Another Redis instance will conflict with FalkorDB. Stop it or change FALKORDB_PORT.
Extension not capturing. Make sure the extension is enabled and the page was loaded after the extension was installed. Reload the AI service tab.
No data on dashboard. Check the extension popup — it shows a count of captured disclosures and connection status. If the count is zero, try sending a message on a supported AI service.
CORS errors. The ingest API includes CORS headers for the extension. If you move the backend to a different origin, update host_permissions in manifest.json.
Docker Compose V2. Use docker compose (no hyphen). Older Linux installations may need docker-compose.
Auth is off by default. Set OMNIGRAPH_PASSWORD in .env to enable dashboard authentication.
Classification requires OpenAI. An OPENAI_API_KEY is required for sensitivity classification. Without it, disclosures are stored but remain unclassified.

Contributing

Contributions are welcome. Please:

Fork the repository
Create a feature branch (git checkout -b feature/my-feature)
Make your changes
Run npm run ci to verify (typecheck + lint + test)
Submit a pull request

Adding a New AI Service

To add support for a new AI service:

Add the service's URL patterns to extension/manifest.json (matches arrays and host_permissions)
Add a service entry in extension/src/services/registry.js with endpoint patterns and a body parser
Optionally add a DOM adapter in extension/src/content/adapters/

License

MIT License. See LICENSE for details.

Name		Name	Last commit message	Last commit date
Latest commit History 20 Commits
.github/workflows		.github/workflows
docs		docs
e2e		e2e
extension		extension
packages/sdk		packages/sdk
public		public
src		src
vscode-extension		vscode-extension
.dockerignore		.dockerignore
.env.example		.env.example
.gitignore		.gitignore
Dockerfile		Dockerfile
LICENSE		LICENSE
README.md		README.md
components.json		components.json
docker-compose.yml		docker-compose.yml
eslint.config.mjs		eslint.config.mjs
next.config.ts		next.config.ts
package-lock.json		package-lock.json
package.json		package.json
playwright.config.ts		playwright.config.ts
postcss.config.mjs		postcss.config.mjs
tsconfig.json		tsconfig.json
vitest.config.ts		vitest.config.ts

Folders and files

Latest commit

History

Repository files navigation

OmniGraph

Why

How It Works

Detection Architecture

Supported AI Services

Quick Start

Prerequisites

1. Clone and install

2. Start FalkorDB

3. Configure environment

4. Start the dev server

5. Load the browser extension

6. Use any AI tool

Environment Variables

Project Structure

Graph Data Model

API Reference

POST /api/ingest

GET /api/graph?action=data&limit=500

GET /api/graph?action=search&query=keyword&limit=20

GET /api/graph?action=timeline&limit=50&offset=0

GET /api/graph?action=dashboard

DELETE /api/graph?id=node-uuid

GET /api/health

Testing

Test Coverage

Build Commands

Tech Stack

Common Gotchas

Contributing

Adding a New AI Service

License

About

Resources

License

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

`POST /api/ingest`

`GET /api/graph?action=data&limit=500`

`GET /api/graph?action=search&query=keyword&limit=20`

`GET /api/graph?action=timeline&limit=50&offset=0`

`GET /api/graph?action=dashboard`

`DELETE /api/graph?id=node-uuid`

`GET /api/health`

Packages