Upload any document. Simulate the future. Predict what happens next.
A multi-agent swarm intelligence engine that simulates public opinion, market sentiment, and social dynamics using GPT-4.1 and in-memory GraphRAG.
Infer is an enterprise-grade multi-agent prediction engine. Upload any document -- a press release, policy draft, financial report, or news article -- and it generates hundreds of AI agents with unique personalities that simulate real-world reactions on social media platforms. Posts, arguments, opinion shifts, sentiment cascades -- all predicted hour by hour.
Document --> Knowledge Graph --> Agent Personas --> Simulation --> Prediction Report
(PDF) (GraphRAG) (GPT-4.1) (Multi-Agent) (Analysis + Chat)
Upload a document and watch Infer extract entities, relationships, and build a live knowledge graph using GraphRAG with Azure OpenAI.
Interactive D3.js knowledge graph with real-time entity extraction -- 138 nodes, 67 relationships, 10 entity types
Infer automatically generates hundreds of AI agent personas, each with unique personalities, opinions, influence levels, memory, and behavioral patterns drawn from the knowledge graph.
75 agent groups, 143 personas total, 14,731 topic associations generated from the document
Each agent has a full biography, social relationship network, behavioral patterns, and memory imprints
Configure and run dual-platform simulations (Twitter + Reddit) with customizable parameters: duration, rounds, peak hours, agent activity periods, and recommendation algorithms.
3-hour simulation across 3 rounds with 143 agents, configurable peak/valley periods and agent configurations
The Report Agent analyzes post-simulation data, conducts agent interviews, searches the knowledge graph, and generates a structured prediction report with sentiment analysis and probability distributions.
AI-generated prediction report: "Simulated Market Futures: S&P 500, Oil Shock, and Fed Dynamics in Early 2026"
Chat directly with the Report Agent or any individual agent from the simulation. Ask them why they said what they said -- full memory and personality persists.
Interactive tools: Report Agent Chat, Insight Forge, Panoramic Tracking, QuickSearch, and Virtual Interviews
Chat with any agent -- big_tech_886, the_fed_209, khameneis_son_369 -- each with their own worldview
+------------------------------------------+
| Infer Frontend |
| Vue 3 + D3.js + Dark Premium UI |
+-------------------+----------------------+
|
REST API
|
+-------------------v----------------------+
| Flask API Layer |
| graph.py | simulation.py | report.py |
+-------------------+----------------------+
|
+-------------------v----------------------+
| Service Layer |
| EntityReader GraphToolsService |
| GraphMemoryUpdater ReportAgent |
+-------------------+----------------------+
|
+-------------------------+-------------------------+
| |
+----------v-----------+ +-----------v-----------+
| In-Memory GraphRAG | | Azure OpenAI |
| +------------------+ | | +------------------+ |
| | Vector Search | | | | GPT-4.1 (LLM) | |
| | Hybrid BM25 | | | | text-embed-3-lg | |
| | JSON Persistence | | | | 3072 dimensions | |
| +------------------+ | | +------------------+ |
+-----------------------+ +-----------------------+
| Feature | Original | Infer |
|---|---|---|
| LLM | Ollama / qwen2.5 (local) | Azure OpenAI GPT-4.1 |
| Embeddings | nomic-embed-text (768d) | text-embedding-3-large (3072d) |
| Graph Database | Neo4j Community Edition | In-Memory GraphRAG (zero dependencies) |
| Setup | Docker + Neo4j + Ollama + GPU | Single pip install + API key |
| Hardware | 16GB RAM + 10GB VRAM minimum | Any machine with internet |
| UI | White/light theme | Premium dark blue/black theme |
| Vector Search | Neo4j vector indexes | Numpy cosine similarity + BM25 hybrid |
| Data Persistence | Neo4j database files | JSON file persistence |
| Use Case | Description |
|---|---|
| PR Crisis Testing | Simulate public reaction to a press release before publishing |
| Trading Signals | Feed financial news, observe simulated market sentiment shifts |
| Policy Impact | Test draft regulations against simulated public response |
| Competitive Analysis | Model how markets react to competitor announcements |
| Product Launch | Predict social media response to product announcements |
| Risk Assessment | Simulate cascading effects of organizational decisions |
- Python 3.11+
- Node.js 18+
- Azure OpenAI API key (or any OpenAI-compatible endpoint)
# Clone the repository
git clone https://github.int.inceptionai.ai/INCEPTION/infer-ai.git
cd infer-ai
# Configure your API key
cp .env.example .env
# Edit .env with your Azure OpenAI credentials
# Install backend dependencies
cd backend
pip install -r requirements.txt
# Install frontend dependencies
cd ../frontend
npm install
# Start the backend
cd ../backend
python run.py
# In another terminal, start the frontend
cd frontend
npm run devOpen http://localhost:3000 -- that's it.
All settings are in .env:
# Azure OpenAI
AZURE_OPENAI_API_KEY=your-key-here
AZURE_OPENAI_ENDPOINT=https://your-endpoint.openai.azure.com/
AZURE_OPENAI_API_VERSION=2024-10-21
AZURE_OPENAI_DEPLOYMENT_NAME=gpt-4.1
# Embeddings
EMBEDDING_MODEL=text-embedding-3-large
EMBEDDING_DIMENSIONS=3072
# Graph Storage (in-memory, no external DB needed)
GRAPH_STORAGE_TYPE=memory
GRAPH_DATA_DIR=./data/graphs| Component | Technology |
|---|---|
| LLM | Azure OpenAI GPT-4.1 |
| Embeddings | text-embedding-3-large (3072d) |
| Knowledge Graph | In-Memory GraphRAG with JSON persistence |
| Search | Hybrid: 0.7x vector cosine + 0.3x BM25 keyword |
| Backend | Python 3.11+ / Flask |
| Frontend | Vue 3 + D3.js + Vite |
| Simulation | OASIS (CAMEL-AI) multi-agent framework |
| NER/RE | GPT-4.1 structured extraction |
infer-ai/
backend/
app/
api/ # Flask REST API endpoints
models/ # Data models (Project, Task)
services/ # Business logic (simulation, reports)
storage/ # GraphRAG storage layer
utils/ # LLM client, logger, file parser
run.py # Backend entry point
frontend/
src/
api/ # Axios API client
components/ # Vue step components
views/ # Page views (Home, Process, Report)
router/ # Vue Router config
index.html # Entry HTML
.env.example # Configuration template
docker-compose.yml # Docker setup (optional)
MIT License. See LICENSE for details.
Built by Anup Roy
INFER -- We don't guess. We infer.