FAQ

How does GPTLint compare to ESLint?

GPTLint is not intended to replace AST-based linting tools like eslint (or biome, ruff, etc).

We absolutely love eslint and strongly recommend that you use an AST-based linter in all of your projects.

The power of GPTLint comes in enforcing higher-level rules that are difficult or impossible to capture with AST-based linters like eslint. GPTLint is therefore intended to augment AST-based linters like eslint to catch potential issues and enforce project-specific standards as early as possible without having to rely solely on PRs from human reviewers.

So what does a higher-level linting rule look like, and when should you use an AST-based linting rule versus an LLM-based linting rule? We answer this in-depth in our rule guidelines, but the short answer is that we recommend using deterministic, AST-based linting wherever possible and augmenting this with LLM-based linting and code reviews to provide a defense-in-depth approach to code quality. LLM-based linting helps to minimize the need for human code reviews, but we believe they are fundamentally complementary as opposed to replacing replacing human code reviews entirely.

How accurate / reliable is gptlint?

See accuracy.

How much will it cost to run gptlint on my codebase?

See cost.

How can I use GPTLint with a custom, local model?

See local models.

How can I use GPTLint with a different LLM provider?

See LLM providers.

How can I disable a rule?

You can disable a rule for a project by adding a config file at the root:

import { recommendedConfig } from 'gptlint'
 
export default [
  ...recommendedConfig,
  {
    rules: {
      'prefer-fetch-over-axios': 'off'
    }
  }
]

How can I disable a rule for a specific file?

Rules can be configured at the file-level using inline comments:

/* gptlint semantic-variable-names: off */

Separate multiple inline configs with commas:

/* gptlint use-esm: off, consistent-identifier-casing: warn */

Or use multiple inline comments:

/* gptlint use-esm: off */
/* gptlint consistent-identifier-casing: warn */

How can I disable linting for a specific file?

Linting can be disabled at the file-level using an inline comment:

/* gptlint-disable */

How can I customize a built-in rule?

Since rules are just markdown files, copy the rule’s markdown file from rules/ into your project and customize it to suit your project’s needs.

You’ll want to disable the original rule and change your custom rule’s name to a project-specific name. Make sure your local config includes your custom rule’s markdown file in its ruleFiles field.

If your change is generally applicable to other projects, consider opening a pull request to GPTLint.

For more guidance around creating and evaluating custom rules that will work well across large codebases as well as expertise on fine-tuning models for use with custom rules, please reach out to our consulting partners.

Are there file size limits?

The following source files will automatically be excluded:

• files matching .gitignore
• files matching an optional .gptlintignore
• binary files
• empty files
• large files over 200kb
• files with over 10k lines
• lines with over 1024 characters will be truncated
• files with more than one line over 1024 characters
• known generated files (package-lock.json, etc)

These filters are similar to github code search’s limits.

What limitations does GPTLint have?

See limitations, cost, and accuracy.

How does privacy work with GPTLint?

GPTLint is a free, open source project which you can run on any codebase.

The only external service GPTLint relies on is an LLM provider, which defaults to OpenAI. OpenAI’s enterprise privacy is to never store, log, or train your code, and their API is fully SOC 2 compliant.

If you need stronger privacy & security guarantees, GPTLint supports using any LLM provider, including local models which can be self-hosted to guarantee that no data is ever transmitted to any external services.

What about fine-tuning?

See our notes on fine-tuning in how it works.

Where can I get support?

For free, open source projects, feel free to DM me @transitive_bs or my co-founder, Scott Silvi, on twitter. Alternatively, open a discussion on this repo if you’re looking for help.

For commercial projects, we’ve partnered with Duality to offer AI consulting services and expertise related to GPTLint. Reach out to our team here, and be sure to include some background info on your project and what you’re looking for.

We are working with several fortune 500 companies as design partners to explore what a productized offering built on top of the GPTLint OSS core might look like. If you’re interested in an enterprise arrangement, please reach out.