AgentExecutor

AgentExecutor is the recommended high-level entry point. It wraps ReActAgent or FunctionCallingAgent behind a consistent interface.

Usage

from synapsekit import AgentExecutor, AgentConfig, CalculatorTool, WebSearchTool
from synapsekit.llm.openai import OpenAILLM
from synapsekit.llm.base import LLMConfig

llm = OpenAILLM(LLMConfig(model="gpt-4o-mini", api_key="sk-..."))

executor = AgentExecutor(AgentConfig(
    llm=llm,
    tools=[CalculatorTool(), WebSearchTool()],
    agent_type="function_calling",  # or "react"
    max_iterations=10,
    system_prompt="You are a helpful research assistant.",
))

# Async
answer = await executor.run("What is the square root of 1764?")

# Sync (scripts / notebooks)
answer = executor.run_sync("What is 12 factorial?")

# Streaming
async for token in executor.stream("Explain the result step by step"):
    print(token, end="")

AgentConfig

Field	Type	Default	Description
llm	BaseLLM	required	LLM provider
tools	list[BaseTool]	required	Available tools
agent_type	"react" | "function_calling"	"react"	Which agent to use
max_iterations	int	10	Max tool-call cycles
system_prompt	str	"You are a helpful AI assistant."	Agent system instruction
verbose	bool	False	Reserved for future logging

Inspecting agent steps

await executor.run("Do some research")

for step in executor.memory.steps:
    print(f"[{step.action}] {step.action_input} → {step.observation}")

Choosing agent type

Situation	Recommended
Using OpenAI or Anthropic	"function_calling"
Using Ollama, Cohere, Mistral, Gemini, Bedrock	"react"
Many tools, need reliable selection	"function_calling"
Want to inspect reasoning chain	"react"