• DATA
  • RANKING
  • INSIGHT
  • HUMAN

How to compare model experiment metrics with your AI

Connect DVC to your AI client to analyze experiment history. Once connected, you can ask your agent to scan through previous runs and pick the winner.

Ask AI about this page

Short answer

How can I compare model experiment metrics using AI?

Your AI uses the DVC Connector to pull metric arrays from your experiment history. It then compares these values to identify which version performed best based on your specific criteria. You get a clear breakdown of the top-performing runs.

The results of your analysis

What you get from the comparison

Running this task gives you these specific outcomes.

DATA

Metric arrays extracted

The AI pulls the specific metric values from your DVC experiments. You get a clean set of numbers to look at.

RANKING

Best version identified

The agent compares the metrics across all available runs. It points you directly to the most successful model version.

INSIGHT

Trend analysis

Your AI looks for patterns in how metrics changed over time. This helps you see if your model is actually improving.

HUMAN

Manual validation required

The AI identifies the winner, but you decide if the results make sense. You remain in control of the final model selection.

The workflow

What your AI does when the data arrives.

The AI handles the heavy lifting of scanning your experiment history.

  1. Fetch experiment history

    The agent scans your DVC repository to find all recorded experiment runs.

    list_experiments

  2. Extract metric values

    It pulls the specific metric arrays associated with each run found in the history.

    list_experiments

  3. Compare performance

    The AI runs a comparison between the different metric sets to find the highest or lowest values.

    list_experiments

  4. Report the winner

    Your agent summarizes which experiment version hit your target metrics.

    list_experiments

Try it

Copy these to start.

Use these prompts to kick off your analysis.

Starting points

These are starting points. Swap out the specific metric names or repo paths to match your setup.

DVC Connector

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

Connector linkhttps://edge.vinkius.com/vk_preview_Myft115Qe8hEZoVFuhiTYRKEW5euUlsOR2XqjK69/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 — DVC capabilities are ready to use.
Configuration · claude_desktop_config.jsonCopy
{
  "mcpServers": {
    "dvc-mcp": {
      "url": "https://edge.vinkius.com/vk_preview_Myft115Qe8hEZoVFuhiTYRKEW5euUlsOR2XqjK69/mcp"
    }
  }
}
Copy into chat04

  • List all my DVC experiments and tell me which one had the highest accuracy.

  • Compare the loss metrics from my last five DVC experiment runs.

  • Find the best performing model in my DVC repo based on the F1 score.

  • Show me a comparison of the metric arrays for the 'production' tagged experiments in DVC.

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

Start here

Connect DVC once, then ask.

Just link your DVC account once. Your credentials stay encrypted, and you can immediately start asking your AI about your experiment data.

Connect DVC to your AI

FAQ

Common questions

  • 01

    Can the AI change my DVC experiment data?

    No. The AI can only read the experiment history to perform comparisons.

  • 02

    Does the AI pick the model to deploy?

    No. The AI identifies the best performing version, but you must make the final decision to deploy it.

  • 03

    Can I compare metrics from different DVC repositories?

    The AI can only access the repositories you have connected via the DVC Connector.

  • 04

    How does the AI know which metric to look for?

    You tell the AI which metric to use in your prompt, such as accuracy or loss.

  • 05

    Can the AI create new experiments for me?

    No. This specific workflow is for reading and comparing existing experiment data.

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

Connect DVC to Claude, Cursor, ChatGPT & more