Production-grade Agentic AI Framework — Multi-Agent Orchestration, Tool-Calling, Memory & Planning
FlashAgent is a modular, production-ready framework for building AI agents that reason, plan, use tools, and collaborate. Part of the FlashVision ecosystem.
| Feature | Description |
|---|---|
| Observe-Think-Act Loop | Core agent loop with structured reasoning |
| Tool Calling | Extensible tool system with web search, code execution, vision, file I/O |
| Memory Systems | Short-term, long-term (vector), and episodic memory |
| Planning Strategies | ReAct, Chain-of-Thought, Tree-of-Thought |
| Multi-Agent Orchestration | Crew-based teams, graph workflows, delegation, consensus |
| LLM Flexibility | OpenAI, HuggingFace local models, routing & fallback |
| Pre-built Agents | Research, Coding, and Vision agents ready to use |
| Benchmarking | Built-in evaluation and performance metrics |
┌─────────────────────────────────────────────────────────┐
│ FlashAgent │
├─────────┬──────────┬──────────┬──────────┬─────────────┤
│ Core │ Tools │ Memory │ Planning │Orchestration│
│ │ │ │ │ │
│ Agent │ Search │ Short │ ReAct │ Crew │
│ Message │ Code │ Long │ CoT │ Graph │
│ State │ File │ Episodic │ ToT │ Delegation │
│ Loop │ Vision │ Embed │ │ Consensus │
│ │ Calc │ │ │ │
├─────────┴──────────┴──────────┴──────────┴─────────────┤
│ LLM Providers │
│ OpenAI │ HuggingFace │ Router │
└─────────────────────────────────────────────────────────┘
git clone https://github.com/FlashVision/FlashVision.git
cd FlashVision/FlashAgent
pip install -e ".[all,dev]"bash setup_env.shpip install -e .pip install -e ".[openai]" # OpenAI support
pip install -e ".[search]" # Web search
pip install -e ".[vision]" # Vision/OCR
pip install -e ".[memory]" # Vector memory (ChromaDB)
pip install -e ".[all]" # Everythingimport asyncio
from flashagent import Agent, ReActPlanner
from flashagent.llm import OpenAILLM
from flashagent.tools import CalculatorTool, WebSearchTool
async def main():
agent = Agent(
name="Assistant",
llm=OpenAILLM(model="gpt-4o"),
tools=[CalculatorTool(), WebSearchTool()],
planner=ReActPlanner(),
max_iterations=10,
)
result = await agent.run("What is 15% of the US GDP?")
print(result)
asyncio.run(main())from flashagent import Agent, Crew
from flashagent.llm import OpenAILLM
llm = OpenAILLM(model="gpt-4o")
crew = Crew(mode="hierarchical")
crew.add_agent(Agent(name="Manager", llm=llm), role="coordinator")
crew.add_agent(Agent(name="Researcher", llm=llm), role="worker")
crew.add_agent(Agent(name="Writer", llm=llm), role="worker")
result = asyncio.run(crew.run("Write a report on AI trends in 2026"))flashagent version
flashagent check
flashagent run --task "Calculate 42 * 17" --model gpt-4o
flashagent chat --model gpt-4o
flashagent benchmark| Page | Description |
|---|---|
| Installation | Setup guide |
| Quick Start | Get running in 5 minutes |
| Agents | Agent architecture deep-dive |
| Tools | Built-in tools & custom tools |
| Memory | Memory systems explained |
| Multi-Agent | Orchestration patterns |
| FAQ | Common questions |
FlashAgent/
├── flashagent/ # Core package
│ ├── core/ # Agent, Message, State, Loop
│ ├── llm/ # LLM providers (OpenAI, HF, Router)
│ ├── tools/ # Tool implementations
│ ├── memory/ # Memory systems
│ ├── planning/ # Planning strategies
│ ├── orchestration/ # Multi-agent coordination
│ ├── solutions/ # Pre-built agent solutions
│ ├── analytics/ # Benchmarking & metrics
│ └── utils/ # Utilities
├── configs/ # YAML configurations
├── examples/ # Usage examples
├── tests/ # Test suite
└── docs/ # Documentation
See CONTRIBUTING.md for guidelines. We welcome contributions!
This project is licensed under the MIT License — see LICENSE for details.
@software{flashagent2026,
title={FlashAgent: Production-grade Agentic AI Framework},
author={FlashVision},
year={2026},
url={https://github.com/FlashVision/FlashVision/tree/main/FlashAgent}
}