Coding assistant for an existing repo

Trying an AI coding tool on a toy project tells you little about using it on a real repository. The questions that matter are about limits, data and review. This guide uses the vendors’ official plan pages. We have not tested these tools, so it does not rank their output.

1. How is usage limited?

Vendors meter heavy use in different ways, and a low monthly price can hide usage-based extras.

• GitHub Copilot lists fixed numbers for Free (2,000 inline completions and 50 chat requests a month), then monthly AI credits on paid plans, where one credit equals $0.01.
• Amazon Q Developer lists 50 agentic requests a month on its Free Tier and a per-line overage of $0.003 for code transformation beyond the Pro allowance.
• Devin says each paid plan has an allowance that refreshes daily and weekly, with extra usage bought at API pricing.
• Cursor draws usage from a pool shared across its products that refreshes weekly, and bills third-party models at API list price from the Premium plan.

2. What happens to your code and prompts?

GitHub says it may use interactions from Copilot Free, Pro and Pro+ to train models unless you opt out, and that Business and Enterprise data is not used. Amazon lists data collection as opt-out on the Free Tier and opted out on Pro. Cursor says that with privacy mode on, code is not used for training by Cursor or its model providers, and does not state the default. For a private repository, find the setting and turn it on before you start.

3. Is IP indemnity included?

GitHub’s plan table lists IP indemnity as not included on individual plans. Amazon lists it as included on Pro and not on the Free Tier. If you ship code commercially, ask whether that matters to your organisation.

4. Does it run where you work?

Copilot lists VS Code, Visual Studio, JetBrains, Eclipse and Xcode, plus the CLI. Amazon Q Developer lists IDE plugins and the CLI. Cursor and Devin are their own editors or environments. A tool that does not fit your editor costs you time before it saves any.

5. Completions or agents?

Completion tools suggest code as you type. Agents take a task and return changes, and Devin describes cloud agents that open pull requests. Either way, you read and run what comes back.

Match the tool to the first question

• Need a familiar editor with a clearly stated free allowance? Start with the GitHub Copilot and Amazon Q Developer comparison, then confirm the plan page before signing up.
• Need an AWS-oriented tool with stated paid-plan IP indemnity? Amazon Q Developer’s page lists it on Pro, not its Free Tier.
• Need an agent to change a repository? Treat the first change as a review exercise, not an autonomous deployment. Keep the change on a branch and run the normal checks yourself.

The directory’s AI coding category collects the documented options and makes the limits visible side by side. It does not measure code quality.

A safe way to try one

This is an illustrative approach, not a test result.

1. Use a branch of a non-critical repository.
2. Turn on the strictest data setting available.
3. Give the tool one small, well-defined task, such as tests for a single function.
4. Review every change and run your test suite.
5. Note how much of your free allowance the task used before you decide to pay.

A trial scorecard

Write down the same five checks for every tool you try: time to set up, allowance used, edits needed, test results, and whether its data setting matched your repository’s needs. A tool that writes more code is not automatically better if it creates more review work.

Related tools

Sources and scope

Plan, limit and data-setting statements above are from the vendors’ official plan pages recorded on 21 October 2026: GitHub Copilot, Amazon Q Developer, Devin and Cursor. Product behaviour changes quickly; re-check the vendor page before committing a repository or paying.