agentblit 0.1.1

Agentic SDK for LLM agents with AgentBlit tools

AgentBlit Python SDK

Build LLM agents with:

• AgentBlit remote tools
• optional local Python tools
• streaming output
• approval-gated actions
• short-term memory with automatic summarization

Install

cd agentblit-python
python -m venv .venv
source .venv/bin/activate   # Windows: .venv\Scripts\activate
pip install -e .

Quick Start

import asyncio
from agentblit import Agent, tool

@tool(description="Add two integers.")
def add(a: int, b: int) -> int:
    return a + b

async def main() -> None:
    agent = Agent(
        model="openai/gpt-4o-mini",
        api_key="YOUR_LLM_KEY",
        agentblit_api_key="YOUR_AGENTBLIT_KEY",
        system_prompt="You are a helpful assistant.",
        custom_tools=[add],
    )

    async for chunk in agent.run("What is 200 + 30? Use tools if needed."):
        print(chunk, end="")
    print()

asyncio.run(main())

---

1) Agent Initialization Configs

Agent(...) accepts:

Key	Type	Required	Default	Values / Format	Use Case
model	str	Yes	-	vendor/model (for example: openai/gpt-4o-mini, anthropic/claude-sonnet-4-0, gemini/gemini-2.0-flash, openrouter/openai/gpt-4o-mini)	Select provider + model
api_key	str	Yes	-	LLM provider API key	Send model requests
agentblit_api_key	str	Yes	-	AgentBlit API key (X-API-Key)	Remote tools + analytics events
agentblit_url	str	No	https://console.agentblit.com	Valid base URL	Self-hosted or staging AgentBlit
system_prompt	str	No	""	Any instruction text	Set assistant behavior
max_history	int	No	5	Integer >= 1	Keep recent messages before summarization
max_tool_rounds	int	No	25	Integer >= 1	Limit LLM-tool loop iterations
debug	bool	No	False	True or False	Enable verbose SDK logs
timeout	float (seconds)	No	30.0	Positive number	Control HTTP timeout for LLM + AgentBlit
approval_callback	async (tool_name, args_dict) -> bool	No	-	Async function	Approve/reject needs_approval tools
custom_tools	list	No	[]	Local @tool callables	Register app-specific local tools

---

2) All SDK Features

• Vendor routing for openai, anthropic, gemini, and openrouter
• Remote AgentBlit tools (/api/tools/list, /api/tools/call)
• Local tools via @tool, custom_tools, and agent.register_tool(...)
• Approval-gated tools (needs_approval) with callback support
• Streaming responses with async for chunk in agent.run(...)
• Multi-round tool calling in one run()
• Memory summarization using max_history
• Automatic analytics batching (agent_init, user_prompt, llm_call, tool_call, tools_updated, agent_loop_error)
• Custom analytics via agent.track(event_type, properties)

---

3) Examples

Example 1: Basic Streaming Agent

agent = Agent(
    model="openai/gpt-4o-mini",
    api_key="YOUR_LLM_KEY",
    agentblit_api_key="YOUR_AGENTBLIT_KEY",
    system_prompt="Be concise and accurate.",
)

Example 2: Approval-Gated Tool

async def approval_callback(tool_name: str, args_dict: dict) -> bool:
    print("Approve tool call?", tool_name, args_dict)
    return tool_name != "delete_production_data"

agent = Agent(
    model="openai/gpt-4o-mini",
    api_key="YOUR_LLM_KEY",
    agentblit_api_key="YOUR_AGENTBLIT_KEY",
    approval_callback=approval_callback,
)

Repo Examples

• examples/basic_agent.py - interactive terminal chat with remote tools.
• examples/custom_tools.py - local @tool functions + AgentBlit tools.