API Reference¶

This section documents the main components of the agents-for-all framework.

class Agent(llm: Model, tools: List[Tool], max_retries: int = 10)¶

Bases: object

Agent class which can ‘do’ actions using llm and tools.

Example code:

from agents_for_all import Agent
from agents_for_all.llms.direct import DirectModel
from agents_for_all.tools.python import Python

llm = DirectModel(
    api_endpoint="http://127.0.0.1:1234/v1/chat/completions",
    model="deepseek-r1-distill-qwen-14b"
)
python = Python()

agent = Agent(llm=llm, tools=[python])
result = agent.do("Create a file with system date as the filename and txt as extension.")
print(result.output) # Final output
print(result.history) # History of steps taken

do(action: str) → AgentResult¶

Do the given action using tools available and the llm specified.

Parameters:: action (str) – The action to perform.
Returns:: Final answer and execution trace.
Return type:: AgentResult

class AgentResult(output: str, history: List[str])¶

Bases: object

Represents the result of running an Agent’s .do() method.

output¶

The final summary response after executing all steps.

Type:: str

history¶

Step-by-step log of what happened during execution.

Type:: List[str]

history: List[str]¶

output: str¶

class Model¶

Bases: ABC

Abstract base class for models or model connectors to be exact. The models are initalized using its subclasses and can be used in agents. Some models are: - DirectModel (with api_endpoint, query_parameter_name, and other_parameters) - OpenAIModel - AntropicModel - GeminiModel

abstractmethod get_response(query: str) → str¶

Get response from the LLM based on the given query.

Parameters:: query (str) – The query the LLM should respond to.
Returns:: The response to the query from the LLM.
Return type:: str

class DirectModel(api_endpoint: str, model: str, parameters: Dict | None = None)¶

Bases: Model

Direct model (or model connector to be exact) which connects to an LLM by using api_endpoint and parameters using OpenAI format (as done by LLMStudio).

Initialize the Direct model.

Parameters:

api_endpoint (str) – The api endpoint to call to get the response.
model (str) – The name of model.
parameters – (Dict, optional): Other parameters to be used while getting responses. Optional.

Returns:

None

get_response(query: str)¶

Get response from the LLM based on the given query.

Parameters:: query (str) – The query the LLM should respond to.
Returns:: The response to the query from the LLM.
Return type:: str

class OpenAIModel(model: str, api_key: str, parameters: Dict | None = None)¶

Bases: Model

OpenAI model connector using the official OpenAI SDK.

Initialize the OpenAI model.

Parameters:

model (str) – The model name (e.g., “gpt-4”, “gpt-3.5-turbo”).
api_key (str) – Your OpenAI API key.
parameters (Dict, optional) – Additional parameters (e.g., temperature).

get_response(query: str) → str¶

Get response from OpenAI chat model.

Parameters:: query (str) – The query the model should respond to.
Returns:: The LLM’s text response.
Return type:: str

class AnthropicModel(model: str, api_key: str, parameters: Dict | None = None)¶

Bases: Model

Anthropic Claude model connector using the official SDK.

Initialize the Anthropic model.

Parameters:

model (str) – The model name (e.g., “claude-3-sonnet-20240229”).
api_key (str) – Your Anthropic API key.
parameters (Dict, optional) – Additional parameters (e.g., temperature).

get_response(query: str) → str¶

Get response from Anthropic model.

Parameters:: query (str) – The query the model should respond to.
Returns:: The model’s response content.
Return type:: str

class GeminiModel(model: str, api_key: str, parameters: Dict | None = None)¶

Bases: Model

Google Gemini model connector using the official Google Generative AI SDK.

Initialize the Gemini model.

Parameters:

model (str) – Gemini model ID (e.g., “gemini-pro”).
api_key (str) – Your Google API key.
parameters (Dict, optional) – Additional generation parameters.

get_response(query: str) → str¶

Get response from Gemini model.

Parameters:: query (str) – The prompt to send.
Returns:: The model’s response content.
Return type:: str

class Tool¶

Bases: ABC

Abstract base class for tools that can be used by agents.

Tools perform tasks or operations when invoked with structured input. Each subclass must implement the execute method and define a description for the LLM or agent to understand the capability of the tool.

abstract property description: str¶

A human-readable description of what the tool does. Should help the LLM choose the appropriate tool.

Returns:: Description of the tool’s purpose and capabilities.
Return type:: str

abstractmethod execute(input_json: Dict) → str¶

Execute the tool’s functionality based on input JSON.

Parameters:: input_json (Dict) – Input parameters to guide tool behavior.
Returns:: Output string describing the result of execution.
Return type:: str

abstract property name: str¶

The name of the tool. Should help the Agent run the appropriate tool.

Returns:: Name of the tool.
Return type:: str

class Python¶

Bases: Tool

A tool that can execute python codes.

Accepts code via input JSON and executes it on the host system using python..

property description: str¶: Explains that this tool executes raw Python code.

execute(input_json: Dict) → str¶

Execute the given code string using python.

Parameters:: input_json (Dict) – Must contain a “code” key with the python command as value.
Returns:: The output or error string from the execution.
Return type:: str

property name: str¶: Python

class Shell¶

Bases: Tool

A tool that can execute shell commands.

Accepts input via input JSON and runs the command on the host shell.

property description: str¶: Executes shell commands on the host system. Input format:

execute(input_json: Dict) → str¶

Execute the tool’s functionality based on input JSON.

Parameters:: input_json (Dict) – Input parameters to guide tool behavior.
Returns:: Output string describing the result of execution.
Return type:: str

property name: str¶: Shell

class File¶

Bases: Tool

A tool that can perform basic file operations: read and write.

Accepts input via input JSON to read from or write to a file.

property description: str¶: Reads from or writes to files.

execute(input_json: Dict) → str¶

Execute the tool’s functionality based on input JSON.

Parameters:: input_json (Dict) – Input parameters to guide tool behavior.
Returns:: Output string describing the result of execution.
Return type:: str

property name: str¶: File

class Math¶

Bases: Tool

A tool to evaluate mathematical expressions safely using sympy.

Accepts an expression string and optional variables to substitute.

property description: str¶: Evaluates symbolic math expressions using sympy. Variables are substituted automatically.

execute(input_json: Dict) → str¶

Execute the tool’s functionality based on input JSON.

Parameters:: input_json (Dict) – Input parameters to guide tool behavior.
Returns:: Output string describing the result of execution.
Return type:: str

property name: str¶: Math

class WebFetcher¶

Bases: Tool

A tool that fetches the content of a web page.

Accepts a URL and returns the of its response body.

property description: str¶: Fetches a webpage using HTTP GET.

execute(input_json: Dict) → str¶

Execute the tool’s functionality based on input JSON.

Parameters:: input_json (Dict) – Input parameters to guide tool behavior.
Returns:: Output string describing the result of execution.
Return type:: str

property name: str¶: WebFetcher

class WebSearch(provider: Literal['google', 'bing'], api_key: str, cx: str = '')¶

Bases: Tool

A tool that performs a web search using Google or Bing API.

Requires API keys during initialization.

Initialize the WebSearch tool.

Parameters:

provider (str) – Either “google” or “bing”.
api_key (str) – API key for the selected provider.
cx (str) – Custom Search Engine ID (only for Google).

Returns:

None

property description: str¶: Performs web search using Google or Bing.

execute(input_json: Dict) → str¶

Execute the tool’s functionality based on input JSON.

Parameters:: input_json (Dict) – Input parameters to guide tool behavior.
Returns:: Output string describing the result of execution.
Return type:: str

property name: str¶: WebSearch

class Email(smtp_host: str, smtp_port: int, username: str, password: str)¶

Bases: Tool

A tool that sends emails via SMTP.

Requires SMTP host, port, and login credentials during initialization.

Initialize the Email tool.

Parameters:

smtp_host (str) – SMTP server hostname (e.g., smtp.gmail.com).
smtp_port (int) – SMTP server port (e.g., 587).
username (str) – Email account username.
password (str) – Email account password or app-specific token.

Returns:

None

property description: str¶: Sends an email using SMTP.

execute(input_json: Dict) → str¶

Execute the tool’s functionality based on input JSON.

Parameters:: input_json (Dict) – Input parameters to guide tool behavior.
Returns:: Output string describing the result of execution.
Return type:: str

property name: str¶: Email

class DataAnalysis¶

Bases: Tool

A tool that runs arbitrary pandas code on a DataFrame created from CSV input.

Example

{: “csv”: “a,bn1,2n3,4”, “code”: “df[‘a’].mean()”

}

Note

Only use safe pandas expressions. This tool does not sandbox or restrict eval().

property description: str¶: Executes pandas code on a DataFrame loaded from CSV. The variable df is automatically defined as the parsed DataFrame.

execute(input_json: Dict) → str¶

Execute the tool’s functionality based on input JSON.

Parameters:: input_json (Dict) – Input parameters to guide tool behavior.
Returns:: Output string describing the result of execution.
Return type:: str

property name: str¶: DataAnalysis