Show HN: SDK for AI Coding with Long Running Tasks

2 points

a year ago

It seems that using Cursor has become really repetitive. My workflow typically goes like this: - Get a task to accomplish on a codebase - Research the web and codebase to create a plan - Send the entire plan to cursor and sit back for it to finish

This flow works really well with a high level of accuracy for the most part and it's becoming obvious that the margin between where it is today and "perfect" is quickly closing.

Because of this flow, I feel that I can replicate it through an SDK that uses a AI cli coder tool with low level control.

In this SDK, you can also use the best coder in the world per: https://aider.chat/docs/leaderboards/

As shown here:

```python from cloudcode import Local import os

def main(): # Use the current directory cwd = os.getcwd() example_file = "example.py"

    # Initialize the Aider SDK in architect mode
    agent = Local(
        working_dir=cwd,
        model="o3",  # Main (planner) model
        editor_model="gpt-4.1",  # Editor model for implementing changes
        architect_mode=True,
        use_git=False,
        api_key=os.getenv("CLOUDCODE_API_KEY")
    )

    # Create or overwrite a simple Python file to modify using the SDK
    agent.create_file(
        example_file,
        """def add(a, b):
    return a + b

""" )

    # Run a coding task using the two-model workflow
    result = agent.code(
        prompt="make this function super cool. just make the math functions better",
        editable_files=[example_file]
    )

    # Print the results
    print("\nTask completed!")
    print(f"Success: {result['success']}")
    print("\nChanges made:")
    print(result["diff"])


    # Display cost information
    print("\nTask completed!")

if __name__ == "__main__": main()

```

Given that you can mimic your workflows from cursor and claude code into this sdk, you shouldn't have to sit around on your computer waiting for the long running task to finish.

That's why I built in a Sandbox class that mounts the ai coder into a remote sandbox directory instead of a local folder so you can deploy an ai coder to the cloud, enabling use cases such as coding from your phone and the browser.

I have already found so many low-handing fruit use cases for this sdk to automate prompting, information gathering, and auto-docs as shown in our github https://github.com/LMSystems-ai/cloud-coding

But I want to take it a step further.

You can get started for free today with this sdk and see our favorable pricing model here: https://docs.cloudcoding.ai/pricing

# Initialize the Aider SDK in architect mode agent = Local( working_dir=cwd, model="o3", # Main (planner) model editor_model="gpt-4.1", # Editor model for implementing changes architect_mode=True, use_git=False, api_key=os.getenv("CLOUDCODE_API_KEY") ) # Create or overwrite a simple Python file to modify using the SDK agent.create_file( example_file, """def add(a, b): return a + b

# Run a coding task using the two-model workflow result = agent.code( prompt="make this function super cool. just make the math functions better", editable_files=[example_file] ) # Print the results print("\nTask completed!") print(f"Success: {result['success']}") print("\nChanges made:") print(result["diff"]) # Display cost information print("\nTask completed!")