An AI coding assistant, with a path to action.

Autocomplete, chat, or agent?

Choose the smallest workflow that can finish the job. A completion is useful when you know the function you need. A conversation helps you reason about an unfamiliar module. An agent is useful when the answer requires reading several files, making a coordinated change, and running a check.

Make the repository part of the question

A generic answer cannot know your dependencies, conventions, or test setup. Open the relevant project and give LoopCode the behavior you want, the entry point if you know it, and a concrete way to check success. In VS Code, add a selected block to the message rather than copying an entire file.

Ask for an explanation before a broad change. LoopCode can search and read the project, then describe the current flow. Use Plan mode when you want to inspect an approach before implementation; Agent mode begins working with the configured permissions.

Explain how a request reaches the checkout handler.
Find the validation and the tests. Do not change files yet.
Identify one place where an invalid quantity could enter the cart.

Turn an answer into a checkable change

An agent can write a plausible test that misses the bug. Read the assertion, confirm it exercises the relevant path, and keep existing coverage. Generated code needs the same review as a contribution from another developer.

1. Define the expected behavior and one failing example.
2. Ask for a focused patch with a regression test.
3. Review the proposed commands and file changes under your permission settings.
4. Check the test output and try the affected behavior yourself before merging.

Local execution is not offline inference

LoopCode operates against the workspace on the machine where its engine runs. Hosted models receive the prompts and relevant code context required to answer. Local-first describes where project tools and conversation storage live; it does not mean that no code leaves the machine or that every model runs offline.

The same LoopCode account and plan are used across the desktop, CLI, and extension. This page describes an agent workflow, not an inline autocomplete product. If you only want keystroke completions, evaluate that capability separately.