Evaluate agent behavior
Run a challenge in the web app, or use your own agent command and verifier with the CLI.
Put agent evaluation where your team works
Evaluate agents in Versalist Cloud or with the CLI on your machine. Customer Google Cloud deployment is planned. On-premises deployment is in development.
Evidence for your next agent change
Choose a deployment path for the way your team tests and reviews agents.
Inspect results and traces
Review scores and failed checks. Hosted trace metadata may be available when capture and access are enabled.
Compare agent changes
Compare local baseline and candidate results with the same verifier before you release a change.
Options
- Versalist CloudAvailableEvaluate agent behavior with a Versalist-operated application on Google Cloud Run in us-central1.See what runs where
- Your Google CloudPlannedA planned deployment option for agent evaluation in a Google Cloud project you own. It is not available yet.See what runs where
- On-premisesIn developmentA Docker Compose package in development aims to put agent evaluation on hosts you control. It is not supported for production use.See what runs where
- Local developmentAvailableRun 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.See what runs where
Compare
| Attribute | Versalist Cloud | Your Google Cloud | On-premises | Local development |
|---|---|---|---|---|
| Status | Available | Planned | In development | Available |
| Operated by | Versalist | You, with Versalist release packages | You | You |
| Application data | Supabase (managed Postgres) | Postgres in your project | Postgres container on your host | .versalist/ in your repository |
| Provider credentials | Google Secret Manager | Secret Manager in your project | Encrypted in your Postgres | Your environment |
| Model calls | Gemini API by default, or providers you add | Vertex AI in your project, or providers you add | Providers you configure; private endpoints require an explicit opt-in | Whatever your agent calls |
| Sandboxed code execution | Not generally available | Not decided | Not available | Your machine |
Where your agent runs is a separate choice
A deployment option decides where the Versalist platform runs. You can evaluate a skill and model on Versalist, or run your own agent and verifier locally with the CLI. Compare where your agent runs.
Works with your harness
Any agent harness you can start from a shell runs under the Versalist CLI, whether you built it on an open-source framework or wrote it yourself. Versalist has no framework-specific adapters. Connect your harness.
Need your own cloud or servers
Tell us your region, data boundaries, identity provider, and models. We confirm what the current release supports before any work begins.
Request a security review