Skip to content

api-reference.md

Latest commit

 

History

History
177 lines (133 loc) · 4.36 KB

api-reference.md

File metadata and controls

177 lines (133 loc) · 4.36 KB

API Reference

Agent

Constructor

Agent(
    llm: BaseLLM,          # LLM provider instance (required)
    identity: str,          # Agent name (required)
    description: str,       # Agent purpose (required)
    task: str = "",         # Default task (used in clan mode)
    verbose: bool = True,   # Print thinking/tool/response output
    tools: list = [],       # List of tool instances
    output_file: str = None,      # Save final result to file
    history_folder: str = ".",    # Conversation history directory
)

Methods

unleash(task: str) -> str

Execute a task. Returns the final answer as a string.

result = agent.unleash(task="Explain quantum computing")

Clan

Constructor

Clan(
    clan_name: str,              # Team name (required)
    manager: Agent,              # Coordinating agent (required)
    members: list[Agent],        # All agents including manager (required)
    shared_instruction: str,     # Instructions for all agents (required)
    goal: str,                   # The task to accomplish (required)
    history_folder: str = "history",  # History directory
    output_file: str = None,          # Save final result to file
)

Methods

unleash()

Plan the work and hand execution to the manager.

clan.unleash()

Built-in Functions (available inside clans)

Function Available To Description
send_message(agent_name, message) All agents Send a task to another agent
ask_user(question) Manager only Ask user a clarifying question

Tool System

@tool Decorator

from unisonai.tools.tool import tool

@tool(name="my_tool", description="What it does")
def my_tool(param: str) -> str:
    return f"Result: {param}"

# Instantiate for use with Agent
agent = Agent(..., tools=[my_tool()])

BaseTool Class

from unisonai.tools.tool import BaseTool, Field
from unisonai.tools.types import ToolParameterType

class MyTool(BaseTool):
    def __init__(self):
        self.name = "my_tool"
        self.description = "What it does"
        self.params = [
            Field(name="param", description="A parameter",
                  field_type=ToolParameterType.STRING, required=True),
        ]
        super().__init__()

    def _run(self, param: str) -> str:
        return f"Result: {param}"

Field

Field(
    name: str,                    # Parameter name
    description: str,             # Description
    field_type: ToolParameterType = STRING,  # Type
    required: bool = True,        # Required?
    default_value: Any = None,    # Default value
)

ToolParameterType

Value Python Type
STRING str
INTEGER int
FLOAT float
BOOLEAN bool
LIST list
DICT dict
ANY any

ToolResult

Returned by tool_instance.run():

Property Type Description
success bool Whether execution succeeded
result Any Return value
error_message str Error message if failed
metadata dict Execution metadata

LLM Providers

Available Providers

Class Module Env Variable
Gemini unisonai.llms GEMINI_API_KEY
Openai unisonai.llms OPENAI_API_KEY
Anthropic unisonai.llms ANTHROPIC_API_KEY
Cohere unisonai.llms COHERE_API_KEY
GroqLLM unisonai.llms GROQ_API_KEY
Mixtral unisonai.llms MIXTRAL_API_KEY
XAILLM unisonai.llms XAI_API_KEY
CerebrasLLM unisonai.llms CEREBRAS_API_KEY

Common LLM Parameters

llm = Gemini(
    model="gemini-2.5-flash",   # Model name
    temperature=0.0,             # Creativity (0.0–1.0)
    max_tokens=2048,             # Max response length
    api_key="your-key",          # API key (or use env var)
)

Custom LLM

from unisonai.llms.Basellm import BaseLLM

class MyLLM(BaseLLM):
    def run(self, prompt: str, save_messages: bool = True) -> str:
        # Your API call here
        ...

    def reset(self):
        self.messages = []