Manage provider credentials in Inference providers. Manage Versalist platform keys in Developer API.
Credential types
- A provider credential authorizes supported calls to OpenAI, Anthropic, Google Gemini, xAI, Mistral, Cohere, DeepSeek, AWS Bedrock, Azure OpenAI, Google Vertex AI, Cloudflare Workers AI, Groq, Cerebras, Together AI, Fireworks AI, Baseten, DeepInfra, Replicate, OpenRouter, Perplexity, or any OpenAI-compatible endpoint you host.
- A Versalist API key authorizes the CLI, MCP server, or supported Versalist API request.
These credentials are not interchangeable.
Add a provider credential
- Open Inference providers.
- Select the provider that the workflow uses.
- Enter the credential fields the provider needs: a single API key for most providers, access key, secret, and region for AWS Bedrock, endpoint URL and key for Azure OpenAI, a service account key file for Google Vertex AI, account ID and token for Cloudflare, or a base URL and optional key for a self-hosted endpoint.
- Pick a model, or enter a custom model id where the provider allows it.
- Save the credential. Versalist verifies the connection right away and shows the result.
- Enable the provider.
Do not add credentials for providers that the workspace does not use.
How credentials are stored
- Secret material is written to Google Secret Manager as its own secret per user and provider, encrypted at rest and versioned on every rotation. Deleting a credential destroys the secret.
- Only the last four characters are kept as a display hint. Keys are never shown again.
- Keys are decrypted in memory for your requests only and scrubbed from error messages.
- Every add, rotation, pause, verification, and delete is recorded in an audit trail.
Provider groups
Frontier labs
OpenAI, Anthropic, Google Gemini, xAI, Mistral AI, Cohere, and DeepSeek authenticate with a single API key.
Enterprise cloud gateways
AWS Bedrock, Azure OpenAI, Google Vertex AI, and Cloudflare Workers AI use your cloud account credentials. Use these when organization infrastructure or policy requires the provider.
High-throughput open weights
Groq, Cerebras, Together AI, Fireworks AI, Baseten, DeepInfra, and Replicate host open models. Replicate support is in beta because its models expose different input schemas.
Routers and self-hosted
OpenRouter and Perplexity use an API key. The Ollama / custom option accepts any OpenAI-compatible base URL that is reachable from Versalist servers over HTTPS.
Capability status
Provider keys: available
Versalist stores supported provider credentials in Google Secret Manager and uses them for supported model calls.
Custom model endpoint: available
Connect vLLM, TGI, LM Studio, Ollama, or a dedicated deployment through the OpenAI-compatible option. Local addresses only work on self-hosted Versalist deployments.
User-managed compute: planned
The current product does not expose a general container, cluster, or graphics processor adapter.
A provider entry in the tool catalog does not confirm an executable integration.
Rotate a credential
- Create a new credential with the provider.
- Replace the stored credential in Versalist. The previous secret version is destroyed.
- Check the verification result, or choose Verify connection from the Manage menu.
- Revoke the old provider credential.
Disable a provider
Disable a provider when you must stop traffic without removing its saved configuration. Remove the credential when the workspace no longer needs it.
Troubleshooting
The CLI rejects authentication
Check the Versalist platform key. Provider credentials do not authenticate CLI requests.
A model request does not reach the provider
Confirm that the correct provider is enabled.
The provider rejects the request
Check credential validity, provider permissions, billing status, and model access.
The failure continues
Record the provider, model, request time, and error message. Then contact support.