Features

Implement

Use Fine to plan and code repository-wide changes, asynchronously.

For developers, the greatest barrier to entry is almost always at the beginning. Fine gives you a jump-start on your coding assignments by drafting a solution and generating boilerplate code - Allowing you to kick-off quickly, and focus on innovation.

Issue in, Pull Request out

The Implement feature follows a structured process to assist you with your coding tasks:

Start with an Issue

Delegating a task to Fine starts by describing the feature or task you want to implement and select the relevant repository.

Pro Fine users can choose which LLM they would like to perform the task. Start with an Issue

Mention files or components in your codebase and issues in your issue management platform (Linear, Jira, GitHub Issues) using the @ symbol. You can also paste links to issues in the conversation.

Tip

Although Fine has access to your entire repository, guiding the Agent with specific context helps deliver better results.

Implementation plan

Fine suggests a detailed, step-by-step implementation plan tailored to your codebase. You can review and edit this plan to ensure it aligns with your goals. Once you're satisfied, click Implement.

Implementation Plan

Implement

Fine works asynchronously in its own development environment to implement the changes, generating code and fleshing out the implementation based on the approved plan.

Implement Changes

View & edit

You'll be able to view the changes made in the code panel on the right hand side. To make further changes, you can choose between requesting from the AI by continuing the conversation, or manually editing the code in the panel on the right. Once you've started making manual edits, a Save button will appear at the bottom of the code panel.

View and edit the code generated by the AI, inside your browser

Rollback

As your conversation with the AI continues, the option to rollback will appear on each previous Implementation task. Clicking rollback will undo all new changes (AI and manual) made, restoring you to the version of the code generated by the PR at that Implementation task.

Undo AI Code changes, reverting to previous versions

Create PR, Review & Revise

When you're satisfied with the AI's code, click "Create PR".

Fine will creates a draft Pull Request for your review, containing the implemented changes. If needed, you can check out the PR to continue working in your IDE or use Fine's integration with GitHub PR to make further revisions, add a summary and review.

AI Sandboxing

Available for Pro

Only available for users in "Pro" plan

Fine operates in a fully-featured cloud development environment, allowing it to:

  • Verify, build, and test the code it generates.
  • Run the code to ensure functionality.
  • Create a feedback loop for continuous improvement of its output.
  • Share the changes with a colleague for external feedback.

This sandboxing approach ensures that the code produced by Fine is of high quality and meets your project's standards before it's presented for review.

Premium users can interact with the sandbox environment via SSH, build the app, test it, and inspect app logs.

Trigger Fine from GitHub or Linear

Triggering from an Issue

  1. Create an issue in your preferred issue tracking system (Linear, Jira, or GitHub).
  2. Comment /code on the issue or add the "Fine" label
  3. Fine will automatically start working on the issue, generating a plan and implementing the changes. Click the "Track Progress" link in Fine's reply to open the web interface where you can continue the conversation.
  4. When finished, Fine will create a PR and provide the link to GitHub for you to review.

Using AI to code new features

Triggering from a Pull Request

Fine can also edit an existing pull request according to review comments:

  1. Open the Pull Request you want Fine to work on.
  2. Leave a review comment that starts with the /revise keyword, followed by your request. You can leave a single comment or multiple comments.
  3. Fine will process your request, make the necessary changes, and push them to the PR's branch.

Using AI for code review

Best practices

  • Provide clear instructions: When assigning issues or requesting revisions, being specific will help Fine understand your requirements.
  • Small-size issues are preferred: Limit the task scope to 1-5 files in the codebase. In general, working in small scopes leaves AI less room for interpretation and guessing.
  • Leverage the planning phase: Take advantage of the step-by-step implementation plan to guide Fine's work and ensure alignment with your project goals.
  • Combine with other Fine features: Use the Implement feature in conjunction with Ask and other capabilities for a comprehensive development workflow. It's useful to chat about your feature first, bringing all the context into the conversation and clarifying the solution approach, and then to trigger an implementation request.
  • Be aware of AI limits: Complex architectural changes or decisions may still require human oversight and intervention.
Previous
The AI Palette