Changelog

Devs can now upload photos to Fine as part of prompts.

Conversations with the agent now include Line Changes Summary. Look out for the familiar green + and red -, indicating how many lines have been added or deleted, next to each step in the Agents’ implementation plans. This can help keep track, making sure the exact changes the Dev wanted were made and only those changes.

We’ve added an actions pane to give Devs quick access to useful tools. Look out for the lightning bolt icon at the top of the AI workspace.

We’ve fixed an issue with scrolling in conversations on mobile.

• Fine now includes pre-built templates the AI can call on for starting new projects from scratch. These templates are designed for those hoping to start with Fine on brand new software, without an existing codebase, and for those who are beginning a totally new part of their project - such as building a frontend to go with your existing backend.

• We’ve updated the initial implementation flow for new tasks without a selected project to automatically begin a new project, call on the relevant template and immediately implement your requests.

• You can now add scripts to New Projects created in Fine, in order to Run them.

• Fixed a bug in context fetching for new projects, ensuring the AI doesn’t try to fetch context from existing projects.

• The stream experience has been updated to show you where the Agent is up to in editing the files.

• When pasting multiple variables in the Environment Variables area of Project Settings, Fine identifies and splits them, propagating them in rows.

• You’re now able to upload a .env file to your projects in Fine.

By popular request, we’ve added the ability to define your own set of custom instructions or rules for the AI Agent. Head to Projects > Project Settings > Instructions.

You can use Project Instructions to set coding style preferences and conventions, include external documentation and style guides.

Instructions will be used by the Agent for every task on the Project, from all users in the workspace.

You can apply different instructions to different projects and if you have shared instructions that you would like to apply to multiple projects, make sure to copy them into each one.

Launched AI Sandboxing

Pro subscribers can now Run the code generated by Fine, enabling you to test and preview the code, all within the browser. No more “it works on my machine”, no more AI code that looks like it works, but doesn’t.

• Try it out - when Fine completes a coding task, hit “Run” in the top right corner of the page to test your code.
• For each project, you'll need to add configuration settings the first time you use the AI Sandbox.
• Fine will run the code in the cloud dev environment dedicated to the conversation - yes, if you have multiple conversations with Fine working at the same time, they can each run in their own virtual machine.
• You can continue the conversation with the AI to make further revisions to the code, based on what you’ve seen in the preview.
  • Describe the issue or change in the conversation box and click enter.
  • Currently, the logs you see in the console are not included in Fine’s context - that is to say, the AI won’t know what has or hasn’t worked unless you tell it. However, you can copy and paste the logs into the conversation to guide the AI further. (We soon will be releasing another update, in which it will happen automatically in case of an error).

Launched Live Previews

When Fine finishes running the code in the Sandbox, you can see what it looks like - both for frontend and backend code.

• At the top of the AI Workspace, next to the "Console" Tab, click "Preview".
• You can also click the Open in New Tab icon to see the Live Preview as a full window.
• Copy and paste the preview URL to get feedback from colleagues or test with tools like Postman.

Access Live Previews from Linear, Slack and GitHub Issues

If you’re using Fine in Linear, Slack or GitHub Issues, clicking “Track Progress” in Fine’s reply will take you to the web interface, where you can use the AI Sandboxing and Live Preview features.

Why this is important for developers

Imagine a world in which AI not only takes your specs and writes the code, but can run and test the code, compare it to the specs and fix itself. Where AI not only writes code that looks right and makes sense to the LLM’s logic, but verifies that it actually works - and meets the exact requirements you defined.

AI Sandboxing is not just another Cloud Dev Environment - it's the next step in achieving this.

With the help of your feedback on the current version of Live Previews, we’ll soon be setting our sights on releasing a complete self-assessing, self-improving AI coding agent - the first ever.

To support new features coming up in the roadmap, we’ve been working on significant infrastructure upgrades and optimizing them to make your AI coding experience faster and more reliable.

• Sped up devcontainer (DC) startup time to 10 seconds

• Introduced prebuilds - ready-to-use devcontainers with your repository and tools - to enable faster AI Sandboxing

• Rewrote DC-server to contribute to speed and stability

• Deprecate DC (socket) proxy

• Improved error handling / alerting for DC service, in cases where devcontainer provisioning fails 

Over the last few weeks, many of our Pro users have accessed the alpha version of our AI Sandboxing and Live Previews features, helping us with feedback and identifying room for improvement. With these new speed and stability improvements, the full version of AI Sandboxing and Live Previews is set to be released soon to all Pro subscribers.

Fine now provisions a virtual dev environment as soon as you start a conversation, instead waiting for you to give an implementation task - so it’s quicker for the AI to get started on coding. Look for the green 🟢 icon at the top right of the screen that indicates the dev environment is live.

Edit the AI code in Fine

When Fine has finished an implementation task, before creating the PR, you can edit the files in the code pane that appears on the right hand side. Remember to click Save, which will appear once you start making manual edits.

This is great for minor fixes and adjustments, saving you the hassle of checking out the branch to your IDE when the AI has written almost exactly what you wanted.

It’s also useful if you get tasks done on the go. Waiting for a bus or have a few minutes in between meetings? Take an outstanding issue, give it to Fine, make any small changes you need and click Create PR - all from your mobile browser.

Rollback Changes in Fine

If you’re having a back-and-forth conversation with the AI, giving instructions, reviewing the results and giving more instructions, sometimes it’s useful to be able to go back a couple of steps. Perhaps you’ve realised that something wasn’t quite necessary or the latest version of the changes wasn’t exactly what you were looking for. It happens to all of us.

Now, when you give further implementation instructions to the Fine AI Agent, the option to rollback appears on the earlier messages. Clicking Rollback will undo all changes since the implementation task you’ve chosen to revert to - both those made by the AI and manual edits.

Start New Projects within Fine

You can now start a new project in Fine, without needing an existing repository in GitHub - or even connecting your account.

To create a new project, head to Projects > Create Project and choose a name. If you’ve connected GitHub, this will create a new repository under the name you’ve chosen.

Alternatively, just start typing your task in Fine’s web interface and if you haven’t chosen a Project to work on, Fine will create a new one.

Smoother delegation of tasks

We’ve improved and standardized the way you interact with Fine from your Issue Management platform, to make it easier to delegate tasks.

In Linear and GitHub Issues, delegate an issue to Fine by adding the “Fine” label or replying /code and you’ll get a notification once Fine has finished and created a PR. If you’ve got a backlog of issues and requests, this is a great way to get ahead of lots of the small tasks that make a difference to your users.

Alternatively, you can comment /guideme and Fine will reply with context and advice on how to approach the task. In particular for bigger tasks, this can help you break it down into steps and help you take on the project strategically.

Lastly, you can comment /ask followed by your question, and Fine will read the comment, search the codebase and its knowledge for information, and provide an answer. No more disturbing the developer sitting opposite you with little questions that break their concentration.

When calling Fine into a ticket, you no longer need to specify which repository you’d like it to work on - Fine understands on its own.

All these features work both in GitHub Issues and in Linear. For all those who don’t use either, we’re surveying which platform you use for Issue Management to help us decide what to integrate next. Click here to vote, it takes just 30 seconds.

Refined GitHub PR features

In GitHub PR, we’ve improved the performance of the /revise and /summary features - they’re now faster and more accurate.

When reviewing a PR, if you need to make a change, comment /revise followed by the change you’d like to make and Fine will do it for you. This saves you pulling the code to your machine to make the edits.

Similarly, if you’ve been handed a PR to review, commenting /summary will give you a breakdown of the PR to help you get started.

Knowledge and Code Panes

We’ve added two new collapsable panes to the Fine web interface. Knowledge allows you to view the sources in your codebase from where Fine drew information and Code allows you to view the changes Fine suggests in implementation tasks.

Integrated o1 & improved model selection

OpenAI announced their new model nearly two weeks ago, promising faster, better coding. We straight away made sure to integrate with Fine, along with the previous OpenAI models and Claude 3.5 Sonnet. The results are fantastic and as before - you don't need your own API keys.

We've also improved the way you select which model to use, so that you can pick the right model for each task.

New Agent Interface

The previously separate “Ask” and “Implement” features are now smarter and work together from a unified interface. Fine can now distinguish whether you’re asking a question or assigning a task and responds with the appropriate workflow. There's no longer a need to use slash commands, just talk to the agent naturally.

We’ve implemented a new backend algorithm that delivers responses more than 10 times faster without compromising quality.

The UI has also been updated so you can see live changes to the codebase as the AI writes them.

Launched Affiliates Program

We launched our brand new affiliate program allowing super-fans and content creators with strong audiences of developers to earn money by promoting Fine. More details can be found here and you can sign up here.

Improved Mobile Experience

We’ve been tweaking and improving our mobile experience to ensure you can get tasks done with the help of AI on the go.

Launched Workflows

Workflows is how you can automate AI agents to perform tasks at a set frequency or following a trigger. Think automatic debugging, documentation, changelog, PR review and more.

Launched Ask

Easily ask Fine any question about your codebase - including your integrated platforms - from the home panel. Find valuable information and key context to make your work easier and more efficient.

Integrated Slack

You can now integrate Slack with Fine and get notifications with agent outputs as part of workflows.

• Fine now supports Claude Opus LLM model. Pro users can configure their agents to use Claude as their backend LLM model. Although slower than the default model, GPT-4 Turbo, Claude Opus has shown to be more accurate.
• AI Chat has gone through a major speed optimization, now responding 2x faster than before.

Assign GitHub Issues to Fine AI Agents

You can now initiate Fine AI agents to work on GitHub issues directly from GitHub.

• Try it out by commenting on the issue with /code command.
• Fine will create a new branch, assign the issue to the AI agent, and create a pull request with the AI agent's changes.

Imporvements

• Chat now supports streaming AI responses.
• Prompt bar now supports multiple lines and rich text formatting (markdown).
• Revamped the AI prompt bar design.
• A few bugs were fixed in the "Issue-to-Pull Request" feature.

Turn Linear Issues into Pull Requests

You can now assign Linear Issues directly to Fine AI agents. To get started, go to Workspace Settings > Integrations and enable the Linear integration.

/code

• To delegate an issue to an AI agent, comment on the issue with /code --repo=owner/repo-name.
• Fine will create a new branch, assign the issue to the AI agent, and create a pull request with the AI agent's changes.

/guideme

Leave a comment with /guideme --reop=owner/repo-name to get a step-by-step guide on how to solve the issue.

Ask around your code

We redesigned AI experience from the ground up, and AI Palette now takes center stage.

The AI Palette lets you chat about around your codebase, and initiate AI Agents to perform repository-wide tasks.

Slash commands

You can interact with AI Palette using slash commands.

• /ask: Lets you talk with an AI that understands your codebase.
• /code: Initiates an AI agent to perform a repository-wide task.

When it comes to visual systems, feedback can also be given while looking at the same monitor, and not at the code. To mimic this live and dynamic interaction between human developers within the Fine system, we are proud to introduce ‘Interactive Preview'.

Today, we are excited to announce the release of Fine’s latest version, carefully crafted with the needs of full stack developers at the forefront. The brand-new version gives users an intuitive environment for software development powered by AI agents. The undeniable highlight being the 'Interactive Preview' feature that fundamentally transforms code development by closing the loop and offering a real-time feedback mechanism for developers.

Feedback is an essential element of software development. This is true for human developers as it is for our AI agents. The more detailed and timely feedback an AI agent can access, the better it becomes at understanding tasks, refining responses, and improving overall function.

Often while working in a team, a developer states their opinion about the piece of code or design another team member created. When it comes to visual systems, feedback can also be given while looking at the same monitor, and not at the code.

To mimic this live and dynamic interaction between human developers within the Fine system, we are proud to introduce ‘Interactive Preview'.

Interactive Preview puts our users in direct control, just like a movie director guiding their actors. Developers can run their code, see the results in real time, and provide feedback directly to the AI agents. It gives developers the option to guide their AI agents as they would guide a team member, helping them understand what works, what doesn’t, and where there's room for improvement. Essentially, forming a closed cycle of task assignment, execution, feedback, and improvement.

Just as real-life feedback sessions bring about clarity, refinement, and consistent improvement, Interactive Preview ensures our AI agents continue advancing in their capabilities. It's a quintessential feature that makes Fine more than a tool - it's a dynamic, adaptive, and collaborative solution, fulfilling the vision of a truly integrated human-AI software development environment.

Tailored For Full Stack Developers

At Fine we are constantly thinking about the end goal: building the virtual developer. Our ‘Interactive Preview' is a testament to our commitment to continuous innovation and giving our users the best possible experience. By offering developers the ability to run their codes and provide live feedback to their AI agents, we're facilitating a more sophisticated, seamless, and interactive programming environment.

Full stack developers, who handle both front-end and back-end processes, require a solution that makes the entire software engineering process more efficient. Now, the AI agents don’t just understand the development context from the provided codebase and documentation, but also take into account the live feedback, becoming even smarter and more efficient.

Think of it as an always available, always eager-to-learn colleague. No matter how many times you run your code, or how many comments you share, the AI agents learn from your feedback, adapting and evolving to better serve you. The more you interact, the more they understand your unique style, your project's specific needs, and the more accurate their solutions become.

Today we are releasing a new version of Fine. In this update we solved many of the issues that users faced in the previous days. If you had trouble setting up a project or running an agent recently, this version might have solved it, so give it a go. We’ve got a new batch of product updates, playbooks, and upcoming events for ya. Here's the latest:

🚀 What's New in Fine

Agents take central stage: We removed the tabs interface and moved the specifications to a panel on the right side of the screen. The main interface will now be dedicated to working with the AI agents: planning, executing, and revising.

Manually stage files for the Agent’s editing: In our latest product update, we introduced mentioning via '@'. This week we introduce manually staging files for the agent to edit. More control, better precision.

Improvements to our DB infrastructure allows importing larger projects: If you previously tried to import a large project there's a good chance the import failed. We now support larger projects, so feel free to try again.

Added GPT-4-32k to our backend: With less limitations on context and rate, the product now works faster, and the quality of suggested tasks and generated code increased.

UX Improvement: We improved the “revision” UX (the pencil button), and added file icons to indicate the file type.

Notebooks can now be deleted: Following multiple requests from the community - Projects and Notebooks can now be deleted. Click the Kebab Menu at the top right to reveal the option.

Multiple Workspaces bugs fixed: Member invitation status now updates, invites are being sent correctly, and the confirmation toast closes correctly.

Graceful error handling: Fine now handles failed tasks with retry mechanisms, and better logging.