CubeLoop — a Pythonic, async-native agent harness core designed for high performance, readability, and production-grade persistence.
Python agent-frameworkagent-harnessagent-runtimeagentic-aiai-agentsai-frameworkanthropicharnessharness-engineeringharness-frameworklanggraphlanggraph-alternativellmmcpopenaipi-agentpi-agent-corepythonpython-agent
"""Resumable tasks — recipe example.
Demonstrates crash-resilient long tasks with checkpointing. Tools are
idempotent (backed by a file-based job store) so re-running after a crash
skips already-completed work.
# Start a job:
uv run python examples/resumable_tasks.py job-1 start
# Kill mid-flight with Ctrl-C, then resume:
uv run python examples/resumable_tasks.py job-1
Requires: cubepi[sqlite]
Set ANTHROPIC_API_KEY or OPENAI_API_KEY before running (see _provider.py).
"""
import asyncio
import json
import os
import sys
from pathlib import Path
from cubeloop import Agent, AgentToolResult, TextContent, tool
from cubeloop.checkpointer import SQLiteCheckpointer
from _provider import MODEL_ID, provider
JOB_DIR = Path(os.environ.get("JOB_DIR", "/tmp/cubepi-jobs"))
JOB_DIR.mkdir(parents=True, exist_ok=True)
@tool(execution_mode="sequential")
async def process_item(item_id: str, *, signal=None) -> AgentToolResult:
"Process a work item. Idempotent — safe to retry after a crash."
job_file = JOB_DIR / f"item-{item_id}.json"
if job_file.exists():
state = json.loads(job_file.read_text())
return AgentToolResult(
content=[TextContent(text=f"Item {item_id} already done: {state['result']}.")],
)
# Simulate slow work (1 second per item).
print(f" [processing item {item_id}...]")
await asyncio.sleep(1)
result = f"processed-{item_id}"
job_file.write_text(json.dumps({"item_id": item_id, "result": result}))
return AgentToolResult(
content=[TextContent(text=f"Item {item_id} done: {result}.")],
)
async def main(thread_id: str, start: bool) -> None:
async with SQLiteCheckpointer("resumable_jobs.db") as cp:
agent = Agent(
model=provider.model(MODEL_ID),
system_prompt=(
"You orchestrate batch processing jobs. "
"Process all items one by one using the process_item tool, "
"then summarize the results."
),
tools=[process_item],
checkpointer=cp,
thread_id=thread_id,
)
def on_event(event, signal=None):
if event.type == "message_update" and event.stream_event.type == "text_delta":
print(event.stream_event.delta, end="", flush=True)
elif event.type == "tool_execution_start":
print(f"\n[tool: {event.tool_name}({event.args})]")
elif event.type == "agent_end":
print()
agent.subscribe(on_event)
if start:
await agent.prompt("Process items A, B, C, and D.")
else:
# Resume: hydrate state from checkpointer then resume.
data = await cp.load(thread_id)
if data is None:
print(f"No saved state for thread {thread_id!r}. Use 'start' to begin.")
return
agent.state.messages = list(data.messages)
last = agent.state.messages[-1] if agent.state.messages else None
last_type = type(last).__name__ if last else "none"
print(f"Resuming thread {thread_id!r} ({len(agent.state.messages)} messages, last={last_type})")
if last_type == "AssistantMessage":
from cubeloop.providers.base import ToolCall
has_pending_tools = any(isinstance(c, ToolCall) for c in last.content)
if not has_pending_tools:
# Run completed normally — no pending tool calls to execute.
# In a real workflow you'd ask the user for the next prompt.
print("Run already completed. Nothing to resume.")
return
# Has unresolved tool calls — resume() will execute them.
await agent.resume()
if __name__ == "__main__":
thread_id = sys.argv[1] if len(sys.argv) > 1 else "job-1"
start = len(sys.argv) > 2 and sys.argv[2] == "start"
asyncio.run(main(thread_id, start))