A High-Performance, Multi-Agent AI System for Startup Decision Intelligence.
AgentAstra transforms generic AI interactions into a structured, multi-model startup war room. By deploying a parallel multi-agent research pipeline, it forces rigorous reasoning, risk evaluation, and actionable output for startup consulting insights. It features a premium dashboard presenting a bull/bear debate system, market radar, and an interactive investor simulator.
Watch the full system in action:
(If the video player does not render, please open the demo.mp4 file directly in your media player).
AgentAstra is built on a high-throughput, asynchronous agentic pipeline using a multi-model routing architecture (GPT-4o, Claude Sonnet, Llama-3.3) for specialized cognitive abilities.
graph TD
UI[Next.js Dashboard] -->|SSE Stream /idea, region, segment| A[Orchestrator Agent]
A -->|Generates Mission Focus| P[Parallel Execution Pool]
P --> M[Market Sizer Agent]
P --> C[Competitor Scout Agent]
P --> PP[Pain Point Agent]
P --> T[Timing Agent]
P --> R[Red Team Agent]
M & C & PP & T & R -->|Aggregated Data| V[Validator Agent]
V --> S[Report Synthesizer]
S --> D[Debate Agent - Bull/Bear/Realist]
D --> UI
UI <--> ST[Shark Tank Investor Simulator]
UI <--> WI[What-If Scenario Simulator]
- Orchestrator Agent: Receives the original idea and dynamically assigns localized parameters to the sub-agents.
- Specialist Agents (Parallelized): 5 specific agents run concurrently using a
ThreadPoolExecutorto unblockasyncioloops. They stream live progress to the frontend via Server-Sent Events (SSE). - Validator Agent: Cross-references the 5 independent outputs to detect contradictions or hallucinations.
- Report Synthesizer: Compiles the verified outputs into a structured startup intelligence report.
- Debate Agent: Takes the synthesized report and synthesizes realistic interactions between an Optimist (Bull), a Pessimist (Bear), and a Realist.
- Framework: Next.js (React 19)
- Styling: Tailwind CSS v4
- Components/Icons: Lucide React, Recharts for Market Radar Scatter Charts
- Communication: Server-Sent Events (SSE) for real-time agent activity feed.
- Framework: FastAPI + Uvicorn
- Streaming:
sse-starlettefor live agent event propagation. - LLM Integrations:
openai,anthropic,groq(Dynamic Model Routing). - Concurrency:
asynciocombined withThreadPoolExecutorfor non-blocking multi-agent inference.
/
├── backend/
│ ├── agents/ # Individual LLM Agent implementations
│ │ ├── orchestrator_agent.py
│ │ ├── market_sizer_agent.py
│ │ ├── ...
│ │ └── shark_tank_agent.py # Investor Simulation
│ ├── utils/ # LLM multi-model routing (llm.py)
│ ├── main.py # FastAPI Application & SSE endpoints
│ └── requirements.txt # Python environment dependencies (Duplicate)
├── frontend/
│ ├── package.json # Next.js dependencies
│ ├── src/ # React UI & Dashboard components
│ └── ...
├── backup/ # Archival code/configurations
├── stratosai/ # Legacy or alternative frontends
├── .env # API Keys (OpenAI, Anthropic, Groq)
├── requirements.txt # Primary Python dependencies
└── demo.mp4 # Full system demonstration video
- Node.js (v20+)
- Python 3.10+
- An
.envfile at the root containing your API Keys:OPENAI_API_KEY=your_key ANTHROPIC_API_KEY=your_key GROQ_API_KEY=your_key
cd backend
python -m venv venv
source venv/bin/activate # Or `venv\Scripts\activate` on Windows
pip install -r ../requirements.txt
uvicorn main:app --reload --port 8000cd frontend
npm install
npm run devThe application will be accessible at http://localhost:3000. Watch the live agent feed as the war room generates your intelligence brief!