AEGIS — Autonomous Enterprise Intelligence Operating System

Bright Data "Web Data UNLOCKED" Hackathon (May 2026) Tracks: Security & Compliance (Primary) // GTM Intelligence // Finance & Market Intelligence Live Console URL: http://localhost:3000 // FastAPI Server URL: http://localhost:8000

---

1. Executive Slogan & Core Vision

AEGIS is an AI-Native Autonomous Enterprise Intelligence Operating System engineered to monitor open-web indices, public subcontractors code repositories, corporate logs, and GTM pricing directories.

AEGIS is not a chatbot. It is an event-driven multi-agent network that bridges the gap between passive reporting and autonomous action.

• PATROL & UNLOCK: Patrols the live web using rotating proxies to extract high-fidelity signals.
• COGNIZE: Maintains semantic profiles in a persistent, evolving relationship knowledge graph.
• SYNTHESIZE: Renders structured executive reports, maps competitor moves, and rates threat risks.
• DISPATCH: Fires active automation pipelines when risk thresholds exceed severity indexes.

The design is engineered to feel like Palantir meets Bloomberg Terminal — high-performance, dark-themed, and cinematic.

---

2. Technical System Architecture

2.1 Data Flow & Services Mapping

                                 +-----------------------+
                                 |  NEXT.JS FRONTEND UI  | <----------+
                                 |  (Command Dashboard)  |            |
                                 +-----------+-----------+            |
                                             |                        | SSE (Server-Sent Events)
                                             | GET /POST              | 
                                             v                        |
                                 +-----------------------+            |
                                 |  FASTAPI WEB SERVER   | -----------+
                                 +-----------+-----------+
                                             |
                                             v (Graph Invoke)
+--------------------------------------------|--------------------------------------------+
|  packages/agents/graph.py                  |                                            |
|                                            v                                            |
|    +---------------------------------------+---------------------------------------+    |
|    |                             SCOUT NODE (Scouts links)                         |    |
|    |  - Runs dynamic queries via Bright Data SERP API                              |    |
|    |  - Streams live EventSource telemetry logs                                    |    |
|    +---------------------------------------+---------------------------------------+    |
|                                            |                                            |
|                                            v (Found URLs)                               |
|    +---------------------------------------+---------------------------------------+    |
|    |                          INVESTIGATE NODE (Crawls & Extracts)                 |    |
|    |  - Scrapes target pages using Bright Data Web Unlocker                        |    |
|    |  - Connects over CDP to Bright Data Scraping Browser for dynamic JS           |    |
|    |  - Mapped entities extracted using AI/ML API (Mistral trials)                 |    |
|    |  - Queries Cognee Semantic Graph to pull historical intelligence profiles    |    |
|    +---------------------------------------+---------------------------------------+    |
|                                            |                                            |
|                                            v (Indicators list)                          |
|    +---------------------------------------+---------------------------------------+    |
|    |                          SYNTHESIZE NODE (Scores & dispatches)                |    |
|    |  - Generates executive report and risk score via AI/ML API (GPT-4o)            |    |
|    |  - Ingests entity profiles persistently into Cognee memory                    |    |
|    |  - Fires HIGH_RISK_DETECTED webhook triggers to TriggerWare                   |    |
|    |  - Commits runs, signals, and nodes to database (PostgreSQL/SQLite)           |    |
|    +-------------------------------------------------------------------------------+    |
+-----------------------------------------------------------------------------------------+

---

3. Deep-Dive Partner Implementations

3.1 Bright Data (SERP API + Web Unlocker + Scraping Browser)

• SERP API Integration (packages/shared/bright_data.py): Leverages your active search credentials. It targets the endpoint https://api.brightdata.com/request on zone aegis_1 and requests raw HTML. A custom regex-based links extractor cleans Google redirection parameters (/url?q=LINK), filters out internal Google resources, and reconstructs organic live URLs and titles dynamically.
• Web Unlocker (packages/shared/bright_data.py): Routes GET requests through the superproxy network (brd.superproxy.io:22225) to bypass Cloudflare, bot barriers, and captchas. If credentials are left blank, it falls back to standard HTTP, and if blocked, injects highly-realistic context-aware mocks (e.g. Anthropic hiring, exposed AWS keys).
• Scraping Browser (packages/shared/bright_data.py): Links Playwright to Bright Data's remote headless Chromium cluster (wss://brd.superproxy.io:9222) over Chrome DevTools Protocol (CDP) to evaluate dynamic client-side JavaScript sites.

3.2 Cognee (Persistent Entity Memory)

• Semantic Graphing (packages/memory/cognee_client.py): Mapped in INVESTIGATE and SYNTHESIZE nodes. Before running a search, Cognee is queried using INSIGHTS searches to recall historical facts. After a run completes, newly discovered entities are added via cognee.add and persistent relationship graphs are compiled via cognee.cognify().
• Dual-Write Fallback: If COGNEE_API_KEY is not provided, the client enters a simulated mode and routes structured nodes and edges to your local database memory_nodes table, guaranteeing that your Memory Graph renders beautifully under any state!

3.3 TriggerWare (Automated Action Pipelines)

• Automated Webhook triggers (packages/workflows/trigger_client.py): When threat scores exceed 7.0/10.0, the Synthesize node dispatches the HIGH_RISK_DETECTED event payload containing your structured report.
• Smart Key Resolver: If TRIGGERWARE_WEBHOOK_URL in .env is populated with a raw webhook key ID (e.g. TjKdmz_...), the trigger client dynamically strips spaces and prefixes it with the standard https://app.triggerware.ai/webhooks/ endpoint, eliminating URL formatting parse failures.

3.4 AI/ML API (Multi-Model Smart Routing & Fallback Trial Engine)

• Smart Model Routing (packages/agents/ai_client.py):
  • Mistral-7b-instruct (Fast / Low Latency): Deployed in the Investigate node to parse raw HTML and structure findings as clean JSON.
  • GPT-4o (Strong Reasoning): Deployed in the Synthesize node to digest all search logs and output the final briefing.
• 404 Model Not Found Fallback Trial Loop: To make entity extraction bulletproof against model name variations or temporary downtime on the AI/ML API, ai_client.py loops through a list of candidate models: ["mistralai/Mistral-7B-Instruct-v0.2", "gpt-4o-mini", "gpt-4o"]. If a model fails or throws a 404, the engine instantly retries with the next candidate in the queue, ensuring zero pipeline crashes.

---

4. Cinematic Command Dashboard Manual

The dashboard interface (http://localhost:3000/dashboard) is divided into distinct operational panels:

1. Top Control Deck: Input box to target companies (e.g. OpenAI, Anthropic) and select scanning focus parameters (Security Risk, Hiring Signal, Pricing Shift) that are passed directly to the LangGraph agents.
2. Cognitive Reasoning Stream (Center-Left): styled as an auto-scrolling hacker terminal. It displays live Server-Sent Events (SSE) progress streams ([SCOUT], [INVESTIGATE], [SYNTHESIZE]) straight from the agent execution loop.
3. Live Threat Beacons (Left): Pulsing radar alert widgets highlighting extracted signals in real time, color-coded by severity (Amber for GTM, Red for security credential exposures).
4. Memory Knowledge Graph (Center-Bottom): An interactive glowing SVG node-edge connection map representing semantic relationships. You can pan, zoom, click nodes to view sidebar property cards, and drag nodes to re-organize the layout.
5. Executive Briefing (Right): The final compiled assessment. Renders a glowing radial threat dial, professional GTM recommendations, scrape source links, and a Download button to save the full report as a structured Markdown file.
6. TriggerWare dispatches (Bottom-Left): Logs triggered automation webhooks and delivery responses in real time.
7. Historical Scans (Bottom-Right): Mapped records pulled directly on mount from your database.

---

5. Setup & Running Instructions

5.1 Environment Configuration (.env)

A prepared .env file is located at your project root. Simply open it and fill in your keys:

# Bright Data Credentials
BRIGHT_DATA_SERP_API_KEY=your_brightdata_serp_api_token
BRIGHT_DATA_SCRAPING_BROWSER_URL=wss://brd-customer-XXXX-zone-XXXX:PASSWORD@brd.superproxy.io:9222
BRIGHT_DATA_WEB_UNLOCKER_URL=  # Leave blank to engage local scraper backup

# AI/ML API Credentials (Redeem your hackathon coupon on aimlapi.com)
AIML_API_KEY=your_aiml_api_key_here
AIML_API_BASE_URL=https://api.aimlapi.com/v1

# Cognee Credentials
COGNEE_API_KEY=  # Leave blank to engage SQLite semantic graph fallback

# TriggerWare Credentials
TRIGGERWARE_WEBHOOK_URL=your_triggerware_webhook_url_or_key_id

# Optional PostgreSQL Database (defaults to a local SQLite 'aegis.db' inside apps/backend if left empty)
DATABASE_URL=
NEXT_PUBLIC_API_URL=http://localhost:8000

5.2 Starting the App Locally (Faster Debugging)

To run the backend and frontend locally using your host Python and Node environments:

Terminal 1 — FastAPI Backend:

cd /Users/apple/Desktop/aegis/apps/backend
pip3 install -r requirements.txt
uvicorn main:app --reload --port 8000

Terminal 2 — Next.js Frontend:

cd /Users/apple/Desktop/aegis/apps/frontend
npm install
npm run dev

Open http://localhost:3000 in your browser to launch the command console!

5.3 Starting the App with Docker Compose

If you prefer running a fully containerized network with PostgreSQL and pgvector:

docker-compose up --build

• Command Dashboard: http://localhost:3000
• FastAPI API Swagger Docs: http://localhost:8000/docs
• Postgres Database: localhost:5432

---

6. Hackathon Judging Tracks Alignment

AEGIS is built to excel across three distinct tracks:

1. Security & Compliance (Primary): Mapped inside the BREACH_SIGNAL_FOUND event parser. Tracks exposed corporate credentials and database tokens on open subcontractor repositories, generates secure response recommendations, and dispatches direct HIGH_RISK_DETECTED dispatches to TriggerWare nodes.
2. GTM Intelligence: Monitors competitive recruitment growths, tracking active roles (e.g. Anthropic's 430 compliance roles surge), locating strategic executive hiring re-allocations, and mapping organizational adjustments.
3. Finance & Market Intelligence: Scrapes competitor blogs and documentation to track strategic pricing model reductions (e.g., Mistral lowering CodeStral API pricing by 20%), alerting finance teams of competitive pricing pivots.