How the agent's working loop fits together
Unlike a code snippet produced in an isolated chat, an agent can use the repository as context and execute project tools. That does not make every answer correct: its assumptions, changes, and evidence still need a developer's review.
- Understand the request and read relevant project files.
- Find the implementation, callers, conventions, and existing tests.
- Prepare an approach or make a focused change under configured permissions.
- Run the checks available in the environment and examine their output.
- Review the diff, explain the outcome, and identify anything still unverified.
Give it a task with a finish line
Useful tasks include explaining a codebase, reproducing a bug, adding a small feature, improving tests, and refactoring a module with a stable contract. Specify expected behavior and the command or interaction that proves it. Ask for codebase understanding before a large change in an unfamiliar project.
Find the cart-total implementation and the existing tests.
Fix the quantity-change bug without changing discounts.
Add a regression test and report the command, result,
and any remaining checks. Keep the patch focused.LoopCode runs where your project lives
The desktop app supports macOS 13 or newer on Apple silicon and Intel, Windows 10 or newer on x64, and Linux x64 installers. The CLI requires Node.js 22.18 or newer. The VS Code extension requires VS Code 1.95 or newer and a trusted workspace; remote editor environments run the engine where the code is.
One LoopCode account covers the app, CLI, and extension, with shared plan usage. Use the desktop browser integration for a reachable web app, terminal tools for repeatable checks, and language-server context when a configured server is available.
Local-first, with explicit model and tool boundaries
Project files and conversation history are stored in the workspace environment. Relevant prompts and code context are sent to hosted models for inference, which requires connectivity. Local-first is not a guarantee of offline operation or that no code leaves the machine.
Agent mode begins working on a request. Plan mode prepares an approach for review before implementation. Choose edit permissions, inspect command output and diffs, and steer or stop active work. Commands follow the configured permission policy.
Start free and compare the workflow
LoopCode is in early access, with Apple notarisation and Windows signing still in progress. The download page explains current installers and first-launch steps. Free is available to start; Pro is ₹1,499 and Max ₹5,999 for 30 days, without automatic renewal. The first payment is refundable within 7 days; see current plan limits and terms on the pricing site.
On Free, LoopCode selects a model for each request. Pro and Max provide model selection. There are no comparative accuracy scores or token-savings benchmarks claimed here: evaluate a representative task with the same repository snapshot and acceptance checks.
KEEP GOING
A useful next step.
Put it to work in your project.
Desktop, CLI, and VS Code. One LoopCode account. Free to start.