Promptary→Frameworks→ReAct

ReAct Prompt Framework: Complete Guide with Examples (2026)

Last updated June 28, 2026 · 1 example · Works with ChatGPT, Claude, Gemini
Quick Answer

ReAct is a six-part prompt framework for AI agents: Role sets the agent persona, Instructions define the goal, Tools lists available tools, Reasoning instructs the agent to think before acting, Action specifies how to call tools, and Observation tells the agent how to process results. Based on the 2022 DeepMind paper ReAct: Synergizing Reasoning and Acting in Language Models.

What Is the ReAct Prompt Framework?

ReAct is designed for agentic AI loops, not single-shot prompts. The Reasoning–Action–Observation cycle instructs the AI to think before each action and incorporate tool results before the next step. In Promptary, it is used to store and version agent system prompts that are fetched at runtime via the REST API.

What Does ReAct Stand For?

Designed for AI agents that use tools and reason step-by-step.

R
Role
The agent's persona and area of expertise.
I
Instructions
The overall goal or mission the agent must accomplish.
T
Tools
List of available tools with descriptions. The agent uses this to decide which tool to call.
Re
Reasoning
Instructions to think out loud before each action — Thought: followed by reasoning.
A
Action
Format for tool calls — Action: [tool name] then Action Input: [parameters].
O
Observation
Instructions for processing tool results before the next Thought step.
+R
Rules (optional)
Termination conditions, final answer format, safety constraints.

When to Use ReAct

Use ReAct for AI agent system prompts where the agent has access to tools (web search, code execution, API calls) and must complete multi-step tasks. For single-shot tasks without tool use, any other framework is simpler and more appropriate.

ReAct Examples

Research agent
Role: You are a research assistant that finds accurate, up-to-date information using web search.
Instructions: Answer the user's question by searching the web, reading relevant pages, and synthesizing a factual answer with sources.
Tools: web_search(query: string) — searches the web and returns result snippets. read_page(url: string) — fetches the full text of a webpage.
Reasoning: Before each action, output 'Thought:' followed by your reasoning about what to search for and why.
Action: Output 'Action: web_search' or 'Action: read_page' followed by 'Action Input:' and the parameters.
Observation: After each tool result, output 'Observation:' followed by a summary of what you learned. Then continue with the next Thought.

ReAct vs Other Frameworks

ReAct vs RISEN: RISEN is for single-shot structured tasks; ReAct is for multi-turn agentic loops with tool use. ReAct should only be used for agent system prompts.

RISENCRITRACEAll frameworks

Use this ReAct template in Promptary — free

Save your first ReAct prompt in the structured editor and get a permanent REST API endpoint. Personal plan is free forever, no credit card required.

Start free

Frequently Asked Questions

What does ReAct stand for in prompt engineering?

In the original DeepMind paper, ReAct stands for Reasoning and Acting. In Promptary's framework fields it maps to Role, Instructions, Tools, Reasoning, Action, and Observation — the structural components of a ReAct agent system prompt.

What is the original ReAct paper?

ReAct: Synergizing Reasoning and Acting in Language Models, published by Shunyu Yao et al. at DeepMind in 2022. The paper showed that interleaving chain-of-thought reasoning with tool actions dramatically improved accuracy on multi-step tasks.

Can I use ReAct prompts with LangChain or LlamaIndex?

Yes. Store the system prompt in Promptary, fetch it via the REST API at runtime, and pass it as the system message to your agent framework.