Crate rune_chain_agent 

ReAct agent loop for rune-chain: LLM reasoning + tool invocation + scratchpad.

rune-chain-agent implements the ReAct (Reasoning + Acting) pattern from the paper ReAct: Synergizing Reasoning and Acting in Language Models. The agent sends the user’s question to an LLM, parses the response for a tool call, executes the tool, feeds the result back as an observation, and repeats until the LLM produces a Final Answer: or the iteration limit is reached.

Features

• AgentExecutor — configurable ReAct loop with any rune_chain_core::Llm backend
• AgentResult — structured output: final answer, iteration count, full scratchpad
• Implements rune_chain_core::Chain so agents compose with sequential and conversational chains
• Tool dispatch via rune_chain_core::Tool — synchronous, zero-allocation tool runs
• Automatic system prompt generation from registered tools
• Configurable max_iterations guard prevents runaway loops

Quick Start

use std::sync::Arc;
use rune_chain_core::{Llm, Tool};
use rune_chain_agent::AgentExecutor;

struct Upper;
impl Tool for Upper {
    fn name(&self) -> &str { "upper" }
    fn description(&self) -> &str { "Converts text to upper case." }
    fn run(&self, input: &str) -> String { input.to_uppercase() }
}

let agent = AgentExecutor::new(llm)
    .tool(Upper)
    .max_iterations(5);

let result = agent.run("Convert 'hello' to upper case.").await?;
println!("Answer: {}", result.output);
println!("Iterations: {}", result.iterations);

Using as a Chain

AgentExecutor implements rune_chain_core::Chain, so it can slot into any chain composition that expects a Chain:

use std::sync::Arc;
use rune_chain_core::{Chain, Llm, prompt_args};
use rune_chain_agent::AgentExecutor;

let agent = AgentExecutor::new(llm);
let output = agent.invoke(prompt_args! { "input" => "What is 2 + 2?" }).await?;
println!("{output}");

ReAct Format

The agent instructs the LLM to follow this scratchpad format:

Thought: <reason about what to do>
Action: <tool_name>
Action Input: <input to the tool>
Observation: <result of the action>
... (repeat as needed)
Thought: I now have enough information.
Final Answer: <your final answer>

Re-exports

pub use agent::AgentExecutor;

pub use agent::AgentResult;

Modules

agent