AgentOps AI Platform

A demo multi-agent AI system built with LangGraph, FastAPI, and Next.js.

Supervisor, Research, Execution, and Evaluator agents collaborate to complete a user goal — with optional semantic memory and observability.

Demo project — designed for local exploration and learning, not production deployment.

---

Features

• Multi-agent orchestration — LangGraph workflow: Supervisor → Research → Execution → Evaluator
• Web search tool — DuckDuckGo search for research context
• Semantic memory — ChromaDB vector store for reusing past high-quality results
• Optional observability — LangSmith + Langfuse tracing (off by default)
• Self-evaluation — quality scoring with retry on failure
• Full-stack demo UI — Next.js frontend + FastAPI backend

---

Architecture

┌─────────────────────────────────────┐
│         Frontend (Next.js)          │
└──────────────────┬──────────────────┘
                   │ HTTP
┌──────────────────┴──────────────────┐
│          Backend (FastAPI)          │
└──────────────────┬──────────────────┘
                   │
         ┌─────────┴─────────┐
         │   LangGraph flow  │
         └─────────┬─────────┘
                   │
     ┌─────────────┼─────────────┐
     ▼             ▼             ▼
 Supervisor    Research      Execution
     │             │             │
     └─────────────┴──────┬──────┘
                          ▼
                      Evaluator

---

Project structure

AgentOps-AI-Platform/
├── backend/                 # FastAPI app + routers
├── frontend/                # Next.js 14 UI
├── src/agentops_ai_platform/
│   ├── agents/              # Supervisor, Research, Execution, Evaluator
│   └── graphs/              # LangGraph workflow
├── memory/                  # JSON + ChromaDB memory layer
├── observability/           # LangSmith / Langfuse helpers
├── tools/                   # Web search tool
├── tests/                   # API smoke tests (CI)
├── scripts/                 # Secret-scan helper
├── .github/workflows/       # CI pipeline
├── .env.example             # Env var template
├── Makefile                 # Local commands
└── requirements.txt

---

Quick start

Prerequisites

• Python 3.10+
• Node.js 18+
• A free Google Gemini API key

Setup

git clone https://github.com/guptaom31619-prog/AgentOps-AI-Platform.git
cd AgentOps-AI-Platform

make venv
source venv/bin/activate

cp .env.example .env.local
# Add your GOOGLE_API_KEY to .env.local

make install
make dev

Service	URL
Frontend	http://localhost:3000
Backend API	http://localhost:8000
Swagger docs	http://localhost:8000/docs

Offline mode (no LLM calls)

# In .env.local
OFFLINE_MODE=1

Useful for UI / API testing without burning API quota.

---

Environment variables

Variable	Required	Description
GOOGLE_API_KEY	Yes	Gemini API key
GEMINI_MODEL	No	Supervisor model (default gemini-2.0-flash)
GEMINI_RESEARCH_MODEL	No	Research agent model
GEMINI_EXECUTION_MODEL	No	Execution agent model
GEMINI_EVALUATOR_MODEL	No	Evaluator agent model
OFFLINE_MODE	No	1 skips live LLM calls
OBSERVABILITY_ENABLED	No	1 enables LangSmith + Langfuse
LANGSMITH_API_KEY	No	LangSmith tracing
LANGSMITH_PROJECT	No	LangSmith project name
LANGFUSE_PUBLIC_KEY / LANGFUSE_SECRET_KEY	No	Langfuse keys
LANGFUSE_HOST	No	Langfuse host URL

See .env.example for the full template.

---

Useful commands

make help      # List all commands
make dev       # Backend + frontend
make backend   # API only (port 8000)
make frontend  # UI only (port 3000)
make test      # Run pytest
make lint      # Ruff + ESLint
make check     # Verify env + deps
make kill      # Free ports 8000 / 3000

---

CI

Every push and pull request to main runs:

1. Security scan — blocks committed secrets
2. Python lint — Ruff
3. TypeScript lint — ESLint
4. Backend tests — pytest (offline mode)
5. Frontend build — next build

Workflow: .github/workflows/ci.yml

---

Tech stack

Layer	Tech
Agents	LangChain, LangGraph, Google Gemini
API	FastAPI, Pydantic, Uvicorn
Memory	ChromaDB
Tools	DuckDuckGo Search
Frontend	Next.js 14, React 18, TypeScript
Observability	LangSmith, Langfuse (optional)

---

License

MIT — see LICENSE.