Skip to content

agent.md

Latest commit

 

History

History
110 lines (79 loc) · 3.03 KB

agent.md

File metadata and controls

110 lines (79 loc) · 3.03 KB

InvokeAgent

The InvokeAgent class is the primary interface between your LLM and real-world APIs. It manages chat history, tools, and agent memory, turning natural-language queries into structured API calls via tool-calling agents.

📦 Import

from invoke_agent.agent import InvokeAgent

🚀 Quickstart

from langchain_openai import ChatOpenAI
from invoke_agent.agent import InvokeAgent

llm = ChatOpenAI(model="gpt-4o")
agent = InvokeAgent(llm, agents=["open-meteo"])

response = agent.chat("What's the weather in Berlin?")
print(response)

  • No agents arg? If you have an agents_map.yaml in your working directory, it will be auto-loaded.

  • Pass explicit file paths or URLs (no {name: path} needed):

    agent = InvokeAgent(llm, agents=[
    "./custom/calendar.json",
    "https://example.com/my_agents.txt",
    "google-calendar"
])

🧠 Constructor

InvokeAgent(
    llm: BaseChatModel,
    agents: Optional[Union[str, List[Union[str, dict]]]] = None,
    tools: Optional[List[BaseTool]] = None,
    context: Optional[str] = None,
    verbose: bool = False,
)

Parameters

Parameter Description
llm A LangChain-compatible chat model (e.g., ChatOpenAI, ChatAnthropic).
agents

  • None (default): loads agents_map.yaml if present, otherwise no integrations.

  • str: path to a YAML map file.

  • List[Union[str, dict]]: each entry may be:

    • A built-in alias ("google-calendar"),
    • A bare file path or URL (infers name from JSON or filename),
    • An explicit mapping { "alias": "path_or_url" }. | | tools | Optional extra tools (defaults to [api_executor]). | | context | Optional custom system prompt; if omitted, uses the internal DEFAULT_CONTEXT. Integrations are still appended afterward. | | verbose | Whether to log intermediate steps (context building, tool calls, etc.). |

🧩 How It Works

  1. Context Assembly Calls build_context() with your context (or default) and agents list/YAML.
  2. Agent Wiring Uses LangChain’s create_tool_calling_agent plus your tools (defaults to api_executor).
  3. Chat History Maintains SystemMessage, HumanMessage, and AIMessage history for multi-turn coherence.
  4. Tool Execution When the LLM emits a function call (api_executor), it performs the HTTP request and feeds back the result.

💬 Chat

response = agent.chat("Book me a meeting at 2pm next Friday.")

Returns the model’s final answer (after any necessary API calls). Each call updates the internal history.

🛠️ Tool Injection

To add custom tools:

from my_custom_tools import my_tool

agent = InvokeAgent(
    llm,
    agents=["open-meteo"],
    tools=[api_executor, my_tool]
)