0.2.0

agentsdk

A lightweight Python SDK for building AI agents with tool use, multi-agent graphs, persistence, and tracing.

Install

pip install agentsdk          # core
pip install agentsdk[otel]    # + OpenTelemetry tracing
pip install agentsdk[dev]     # + pytest / dotenv

Quickstart

import asyncio, os
from agentsdk import Agent, AgentConfig, GroqProvider

agent = Agent(
    config=AgentConfig(
        name="MyAgent",
        system_prompt="You are a helpful assistant.",
    ),
    llm=GroqProvider(api_key=os.environ["GROQ_API_KEY"]),
)

async def main():
    result = await agent.run("What is the capital of France?")
    print(result.output)

asyncio.run(main())

Tools

from agentsdk import tool, ToolRegistry, Agent, AgentConfig, GroqProvider

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

registry = ToolRegistry()
registry.register(add)

agent = Agent(config=AgentConfig(name="Calc", system_prompt="Use tools."),
              llm=GroqProvider(...), registry=registry)

Multi-agent Graph

from agentsdk import AgentNode, Edge, AgentGraph, GraphRunner

graph = AgentGraph()
graph.add_node(AgentNode("researcher", researcher_agent))
graph.add_node(AgentNode("writer", writer_agent))
graph.add_edge(Edge("researcher", "writer", data_map={"output": "input"}))
graph.set_entry("researcher"); graph.set_exit("writer")
result = await GraphRunner(graph).run({"input": "Explain black holes"})

Persistence

from agentsdk import FileCheckpointStore, SessionManager, Agent

store = FileCheckpointStore(base_dir=".agentsdk/checkpoints")
session_mgr = SessionManager(store=store, agent_name="MyAgent")
agent = Agent(config=..., llm=..., session_manager=session_mgr)

# History is saved and reloaded automatically across runs:
await agent.run("My favourite language is Python.", session_id="user-001")
await agent.run("What language did I mention?",     session_id="user-001")

# Fork a session to branch an agent run:
forked = await session_mgr.fork("user-001", "user-001-branch")

Tracing (requires agentsdk[otel])

from agentsdk.observability import SDKTracer, TracedLLMProvider, TracedAgent, print_trace

tracer     = SDKTracer(service_name="myapp")
traced_llm = TracedLLMProvider(provider=GroqProvider(...), tracer=tracer)
agent      = TracedAgent(config=..., llm=traced_llm, tracer=tracer)

result, ctx = await agent.run("Summarise the last quarter earnings.")
print_trace(ctx)
# ╔══ Trace: MyAgent ══════════════════════
# ║  Session     : (none)
# ║  Trace ID    : 6744d6eca33853c5bba0...
# ║  Duration    : 3680ms
# ║  LLM calls   : 2
# ║  Tool calls  : 1
# ║  Tokens      : 1292 in / 36 out
# ╚═════════════════════════════════════════

CLI

# Scaffold a new agent project
scaffold-agent new myproject

# Interactive REPL against any agent file
scaffold-agent run myproject/agents/main.py

# Inspect a saved checkpoint
scaffold-agent trace .agentsdk/checkpoints/MyAgent/user-001.json

# List all sessions for an agent
scaffold-agent list-sessions MyAgent