• RUNNING
  • DEPLOYED
  • MONITOR
  • HUMAN

How to fine-tune custom models with your AI

Connect Replicate to your AI client to manage training jobs. Once connected, you can start new sessions and track their status directly in chat.

Ask AI about this page

Short answer

How can I fine-tune models using my AI?

You tell your AI to start a training job using your datasets. It handles the execution on Replicate and monitors the progress until the model is ready. You'll see the results of these training runs through the following breakdown.

Training outcomes

What you get from training

These are the results of your training requests.

RUNNING

Active training jobs

The AI starts the process and keeps an eye on the logs. You get updates on whether the job is still working or if it failed.

DEPLOYED

Ready to use models

Once a job finishes, the AI creates a new model version. This makes your custom weights available for immediate use.

MONITOR

Real-time status

Your agent can check on any existing training run. You'll know exactly when a model is ready without checking a separate dashboard.

HUMAN

Dataset management

The AI cannot prepare or clean your training data. You must host your files in a place Replicate can reach before you ask the AI to start.

The workflow

What your AI does when you request training.

The AI manages the heavy lifting on Replicate so you can stay in your chat client.

  1. Initiate training

    The AI sends your training configuration and dataset location to Replicate to kick off a new job.

    create_training

  2. Track progress

    The AI checks the current status of your training run to see if it's still active or completed.

    get_training

  3. Register model

    When the training succeeds, the AI creates a new model version from the resulting weights.

    create_model

  4. Verify status

    The AI checks the training details again to confirm everything finished correctly.

    get_training

Try it

Copy these to start.

Use these prompts to begin your fine-tuning work.

Starting points

Swap out the dataset URLs and model names for your own specific files.

Replicate Connector

You're all set. Choose your MCP client and follow the setup instructions.

Connector linkhttps://edge.vinkius.com/vk_preview_ULeVbcRjoLCkIr9x93q2HML8VJd4AXmk4CWHHPso/mcp

Claude Desktop

Follow the steps below to connect in seconds.

  1. 1In Claude Desktop, open Settings → Connectors.
  2. 2Click “Add custom connector” and paste the connector link above as the remote MCP server URL.
  3. 3Click Add and start a new chat — Replicate capabilities are ready to use.
Configuration · claude_desktop_config.jsonCopy
{
  "mcpServers": {
    "replicate-extended-mcp": {
      "url": "https://edge.vinkius.com/vk_preview_ULeVbcRjoLCkIr9x93q2HML8VJd4AXmk4CWHHPso/mcp"
    }
  }
}
Copy into chat04

  • Start a new training session using the dataset at https://example.com/my-data.jsonl

  • Check the status of my training job to see if it is done.

  • Create a new model from the training run I just finished.

  • Get the status of the training job with ID training_12345.

  • Claude
  • ChatGPT
  • Cursor
  • VS Code
  • Windsurf
  • Claude Code
  • JetBrains
  • Cline

Start here

Connect Replicate once, then ask.

Link your Replicate account to Vinkius. Your credentials stay encrypted, and you can start managing models immediately.

Connect Replicate to your AI

FAQ

How this task behaves

  • 01

    Can the AI clean my training data?

    No. The AI can only start training once you provide a valid dataset URL. It cannot edit your files.

  • 02

    Can the AI delete my models?

    No. This MCP only has tools to create training jobs and models. It cannot delete anything.

  • 03

    Can I see if a training job failed?

    Yes. The AI can check the status of your training runs and report back if there was an error.

  • 04

    Does the AI handle my Replicate billing?

    No. All costs are handled through your own Replicate account.

  • 05

    Can I stop a training session?

    No. There is no tool available to cancel or stop a training job once it has started.

  • More questions about Replicate? The Connector page answers them. See everything the Replicate Connector can do

Connect Replicate to Claude, Cursor, ChatGPT & more