Skip to main content

Local development

Run your agent and its checks on your own machine with the Versalist CLI. Records stay in your repository. Submit sends a project URL and metadata.

Run your agent locallyCompare deployment options

Agent evaluation and evidence

Run your agent command and verifier on your machine. Compare baseline and candidate results with the same verifier. Local records are not hosted traces and Versalist does not verify them.

Explore results and traces

What runs where

Your agent
Your machine, started by versalist run --command
Checks
Your machine, started by versalist evaluate --command
Baseline and candidate comparison
Your machine. versalist compare exits non-zero when the candidate regresses.
Run records and logs
.versalist/ in your repository
Challenge files and public cases
Downloaded from Versalist by versalist start
Submissions
Sent to Versalist only when you run versalist submit: the project URL and metadata

You provide

  • Node.js 18.17 or later.
  • A Versalist API key, for start and submit.
  • The command that starts your agent, and a command that checks its output.

Versalist provides

  • The @versalist/cli package on npm.
  • Challenge files and public test cases.
  • An MCP server for coding agents: versalist mcp.

Current limits

  • Versalist does not verify local records. They are evidence for you and your reviewers.
  • Challenge files include public cases only.
  • Running an Ollama model with platform judging is not in the published CLI yet.

How to start

  1. 1Load a challengenpx -y @versalist/cli start <challenge> writes the challenge files to your repository.
  2. 2Record two runsversalist run --command "<your agent>" --label baseline, then again for the candidate.
  3. 3Check and compareversalist evaluate --command "<your check>" for each run, then versalist compare.
Run your agent locally

Other deployment options