agentic-sdk

⚠️ Pre-release placeholder. This package reserves the name — the real thing is coming soon.

Python bindings for the Agentic C++ SDK — a unified library for orchestrating AI coding agents and direct LLM APIs.

What it is

Agentic is a C++ SDK (with Python bindings via this package) that treats coding agents and LLM conversations as first-class sessions with a shared event model:

• Unified session API — the same prompt / event-stream interface whether you're talking to an LLM directly or driving a CLI coding agent like Claude Code, Gemini CLI, or Codex.
• Structured event database — every tool call, reasoning step, and response is captured in a queryable local database. Slice by session, tool, time, or full-text search.
• Pythonic tool definitions — define tools as decorated functions with Pydantic models. The LLM gets the schema, you get the type safety.
• Async event streams — subscribe to events as they happen, across sessions, across processes.
• Forking — branch a session at any point to explore alternate paths without losing the original context.
• Extensible — providers for new coding agents or LLMs load as dynamic libraries.

Status

Not yet functional. The C++ SDK is under active development and the Python bindings will land shortly after it stabilizes.

If you're curious about the design, the full interface design document lives in the main repo.

Coming soon

import agentic
from agentic.tools import tool
from pydantic import BaseModel

class WeatherArgs(BaseModel):
    city: str

@tool(description="Get the weather for a city")
def weather(args: WeatherArgs) -> dict:
    return {"temp": 72, "conditions": "sunny"}

ctx = agentic.Context()

with ctx.llm_session(model="anthropic/claude-sonnet-4", tools=[weather]) as session:
    async for event in session.prompt("What's the weather in Portland?"):
        print(f"{event.event_type}: {event.content}")

License

0BSD — BSD Zero Clause. Do whatever you want.