Direct answer: Refact.ai supports fine-tuning on your own repositories by building a training set from your code and running a local training job; the payoff is completions that reference internal APIs, but it takes real GPU hours and only helps on code the model hasn’t seen.

When is fine-tuning worth it?

When your codebase leans on internal frameworks, domain-specific naming, or a monorepo structure that general models fumble. A fintech repo with a custom rules engine, for example, benefits more than a standard React app that popular models already handle well. If your stack looks like everyone else’s, you’ll see less lift.

What data does it need?

Your repositories, ideally clean ones. Point it at the repos you want it to learn and let it extract training examples. Two practical filters:

• Exclude generated files, vendored dependencies, and minified code.
• Skip repos with heavy copy-paste or abandoned experiments — you’ll teach the model your worst habits.

Curating what goes in matters more than volume. Garbage in, confident garbage out.

How do I run training?

1. Connect the source repos to your Refact.ai server.
2. Let it build the dataset from commit history and file structure.
3. Choose a base model that fits your GPU memory.
4. Run the training job and monitor loss.
5. Deploy the tuned model and point your IDE plugin at it.

Expect this to take hours on a single GPU, not minutes. Plan a maintenance window.

How do I know if it actually helped?

Measure before and after on a real task, not vibes. Keep a set of internal files the model hasn’t trained on directly, and compare completion acceptance rates. If your team accepts tuned suggestions more often — and crucially, if they stop correcting internal API names — the training paid off. If not, you’ve spent GPU time for nothing and should revert to the general model.

What are the risks?

Overfitting to old patterns is the big one. Train on a codebase mid-migration and you’ll get fluent suggestions for the framework you’re leaving. There’s also staleness: a model trained last quarter won’t know the API you added last week, so you need a retraining cadence, which keeps adding cost.

Key Takeaways

• Fine-tuning pays off most on internal frameworks and unusual conventions.
• Curate training repos — exclude generated, vendored, and abandoned code.
• Training takes hours on a single GPU, not minutes.
• Measure completion acceptance rates before and after to judge value.
• Models go stale; budget for periodic retraining.

Direct answer: Refact.ai is an open-source AI coding assistant available as a cloud service or a self-hosted server you run on your own hardware, supporting IDE plugins for VS Code and JetBrains; you can point it at a local or your own hosted model for full privacy.

Why self-host a coding assistant?

Compliance and secrets. If your repo contains regulated data or client code you can’t send to a third-party API, a self-hosted assistant is often the only way to use one at all. Refact.ai also lets you fine-tune on your own codebase, which can improve completions on internal frameworks a general model has never seen.

How do I install it?

1. Install Docker on a machine with a GPU if you want local inference (a 24GB card handles mid-size models).
2. Pull the Refact.ai server image and start it.
3. Create an admin account on first run.
4. Install the IDE plugin — VS Code or JetBrains — and point it at your server URL.
5. Log in from the IDE with the account you created.

Once connected, completions and chat route through your server, not the cloud.

Which model should I run?

You can also connect an external API if you’re allowed to send prompts out — a middle ground that keeps the tool on-prem while the model is hosted.

How do I fine-tune it on my codebase?

Refact.ai can train on your repositories so completions pick up your naming and patterns. It’s not instant — you point it at repos, it builds a dataset, and you run training. For a large monorepo with unusual conventions, this is where the tool pulls ahead of generic assistants that keep suggesting idioms your team abandoned.

What are the costs and limits?

The software is open source; your cost is hardware and the time to maintain it. Running inference on a GPU 24/7 isn’t free — a mid-tier cloud GPU is hundreds of dollars a month. Small local models also trail frontier models on hard reasoning, so expect weaker results on complicated refactors even if autocomplete feels fine.

Who is Refact.ai for?

Teams with strict data rules and a GPU to spare, and developers who want autocomplete tuned to their own code. If you can use cloud assistants freely, a managed tool needs less babysitting.

Key Takeaways

• Refact.ai is open source and self-hostable, with VS Code and JetBrains plugins.
• A 24GB GPU handles mid-size quantized models; CPU-only is for testing.
• You can fine-tune on internal repos to match your conventions.
• Total cost is hardware plus maintenance, not a subscription.
• Local models trail frontier models on complex reasoning tasks.