Agentic Workflow System

A platform for building, running, and interacting with agentic workflows powered by Large Language Models (LLMs). This project provides a modular backend for orchestrating agent reasoning, planning, and tool use, along with a modern React frontend for interactive user experiences.

---

Features

• Agentic Reasoning: Agents can plan, decompose, and execute complex tasks using a set of modular tools.
• LLM Integration: Supports local, HuggingFace, and OpenAI LLMs for planning, summarization, coding, and more.
• Pluggable Tools: Summarization, citation generation, open-access journal search, and code generation.
• Memory System: Short-term and long-term memory for contextual, stateful agent behavior.
• Modern Frontend: React + Vite UI for submitting queries, viewing plans, and interacting with agent results.
• API-First: FastAPI backend with REST endpoints for agent queries and results.

---

Architecture

┌────────────┐      HTTP API     ┌──────────────┐
│  Frontend  │ ◄───────────────► │   Backend    │
│  (React)   │                   │  (FastAPI)   │
└────────────┘ ◄───────────────► │   Agentic    │
								 │   Workflow   │
								 └──────────────┘

• frontend/: React app (Vite, TypeScript) for user interaction
• backend/: FastAPI app with agent core, planning, memory, and tool modules

---

Quickstart

1. Backend Setup

cd backend
pip install -r requirements.txt
# Configure environment variables in a .env file (see below)
uvicorn src.main:app --reload
# App runs at http://localhost:8000

Environment Variables (.env)

# Example for local LLM
LLM_CLIENT_TYPE=local
LLM_API_URL=http://localhost:1234/api/v1/chat
LLM_MODEL=qwen2.5-coder-32b-instruct

# Or for OpenAI
# LLM_CLIENT_TYPE=openai
# OPENAI_API_KEY=sk-...
# OPENAI_MODEL=gpt-3.5-turbo

2. Frontend Setup

cd frontend
npm install
npm run dev
# App runs at http://localhost:3000

---

Usage

1. Start the backend and frontend servers.
2. Open the frontend in your browser.
3. Enter a query and select an agent role (e.g., Researcher, Coder).
4. The agent will plan, execute, and display results, including reasoning steps and memory.

---

Backend Overview

• core/agent.py: Main agent class (planning, memory, tool use)
• core/planner.py: Planner interface and LLM-based planner
• core/tools.py: Modular tools (summarization, citation, search, coding, formatting)
• core/memory.py: Short-term and long-term memory system
• core/clients/: LLM client interfaces (local, HuggingFace, OpenAI)
• src/routes/: FastAPI routes for agent query and result endpoints

Frontend Overview

• src/pages/App.tsx: Main app page
• src/components/AgentInteraction.tsx: Query form and result display
• src/api/agent.ts: API client for backend

---