Skip to main content
Documentation menu / Product directory

Product directory

Find products and compare them against task, infrastructure, and cost requirements.

The public product directory contains model providers, frameworks, agents, evaluation tools, and infrastructure products. Directory entries do not grant provider access.

Some products expose tools: functions an agent can call, through an HTTP API, MCP server, SDK, or CLI. Where a curator has reviewed them, a product page links to its tool contracts. Versalist Tools explains what a contract records.

Find a product

  1. Define the task before you select a product.
  2. Open the product directory.
  3. Filter by category or capability.
  4. Open a product detail page.
  5. Review related challenges, prompts, and guides.

Compare products

Use the same requirements for each candidate.

  • Task support: Confirm that the product performs the required task.
  • Evidence: Identify the logs, outputs, or metrics that the product produces.
  • Hosting: Record where the product processes and stores data.
  • Access: Check authentication, provider keys, and organization policy.
  • Cost: Compare cost at the expected request and token volume.
  • Latency: Measure response time under the expected workload.

Test candidates against one challenge. Do not compare products with different tasks or evaluators.

Model providers

The directory includes providers such as OpenAI, Anthropic, Google, Azure, and Amazon Bedrock. It also includes specialized providers.

A directory entry describes a provider. It does not confirm that Versalist can execute a model through that provider.

Provider credentials

Add a credential in Inference providers only when a Versalist workflow must call that provider.

  1. Select the provider that the workflow uses.
  2. Store the provider credential.
  3. Enable the provider.
  4. Run a small test request.
  5. Disable or rotate the credential when the environment changes.

Provider credentials are different from Versalist API keys. Read API keys before you configure CLI authentication.

Use a product in an evaluation

  1. Record the product and version.
  2. Record the model and model version when applicable.
  3. Keep the challenge and evaluator unchanged.
  4. Run the baseline.
  5. Replace one product or configuration value.
  6. Run the candidate.
  7. Compare results and available bounded trace metadata.

Current Episode traces do not record tool execution.

  • Challenges explains task environments and evaluation rules.
  • Integrations explains provider credentials and current capability limits.
  • Prompt library contains prompt examples linked to products and challenges.
Was this page helpful?