cowork-to-code-bridge — delegate to Claude Code on local machine from LlamaIndex agents

[abhinaykrupa]

Hey LlamaIndex community!

Building LlamaIndex agents or workflows that need to execute code or tasks on the user's actual local machine (not a cloud sandbox)? Here's a pattern that may be useful.

What it is

cowork-to-code-bridge — an async file-based bridge that lets LlamaIndex agents delegate complex execution tasks to Claude Code on the local machine. The pattern fits naturally into LlamaIndex's tool/agent model.

Integration with LlamaIndex

from llama_index.core.tools import FunctionTool
from cowork_to_code_bridge.client import queue_task, poll_task_result
import time

def execute_locally(prompt: str) -> str:
    """Execute a complex coding task on the local machine via Claude Code."""
    task_id = queue_task(prompt, idempotency_key=f"llama-{hash(prompt)}")
    while True:
        result = poll_task_result(task_id)
        if result["status"] in ("completed", "cancelled", "failed"):
            return result.get("output", result.get("error", ""))
        time.sleep(2)

# Add to your agent
local_executor_tool = FunctionTool.from_defaults(fn=execute_locally)

What this unlocks

• Full local context — user's repos, shell, installed tools, environment
• Async execution — fire-and-forget with queue_task, poll with poll_task_result
• Idempotent — safe to retry; cached by idempotency_key
• Budget control — max_budget_usd cap per operation
• Loop detection — circuit-breaker for runaway tool calls

Why async matters for agent frameworks

LlamaIndex workflows can block on long-running tasks. The async model lets your agent queue work, do other things, then collect results — no thread blocking.

Repo: https://github.com/abhinaykrupa/cowork-to-code-bridge

Happy to write a LlamaIndex-specific integration example if there's interest — especially around the AgentRunner or SubQuestionQueryEngine patterns where local execution could be one of the sub-tools.