Can I test ModelRiver without real provider keys?
Yes. Create a workflow with a structured output and no provider credentials. ModelRiver can synthesise schema-compliant fallback responses for testing.
Do you support streaming responses?
Yes. The ModelRiver client SDK (@modelriver/client) streams responses in real time over WebSocket. Your backend starts the request and returns a one-time token, then the frontend connects and receives the reply as it is generated. See the client SDK docs and the streaming guide.
How many workflows can I create?
As many as you need. Use clear naming conventions and organise by team or feature for easy discovery.
How do I import or export workflows?
ModelRiver supports portable version 1 bundles. Export a project from the Workflows or Structured outputs section, and import the JSON file into any project. Everything is created atomically and existing records are never overwritten. See Import & export.
What are workflow templates?
Ready-made workflow bundles for common business tasks - refund handling, ticket triage, lead qualification, and more. Download one from the templates page, import it into your project, and follow the integration guides to connect it to your app.
Can workflows call my backend?
Yes. Workflows can run a backend pipeline: ModelRiver delivers events to your webhook URL, your backend verifies or enriches the data, and sends the result back through a callback - all without exposing AI credentials to the frontend. See Backend pipeline.
How should I separate environments?
Create distinct projects (for example, "Production" and "Staging") and issue dedicated API keys per environment. This keeps credentials, logs, and analytics isolated.
What data does ModelRiver store?
Request payloads, responses, cached data, and token usage metrics are stored for observability and troubleshooting. Purge logs as required by your retention policy.
How do I rotate provider credentials?
Update the credential under Settings → Providers. ModelRiver encrypts the new value immediately.
Can I use my own provider API keys?
Yes. Connect your own keys for OpenAI, Anthropic, Google, and other providers in the Providers section. You pay providers directly, and ModelRiver adds routing, failover, budgets, and observability on top.
How do I control AI spending?
Set project and environment spending limits so runaway tests or traffic spikes stop before they turn into surprise invoices. See Spending limits for setup and monitoring.
Are webhooks signed?
Yes. Every delivery includes an mr-signature header - an HMAC-SHA256 of the raw request body using your webhook secret - plus a timestamp and webhook ID so you can verify the request before acting on it. See Webhooks.
Is the API OpenAI-compatible?
Yes. ModelRiver exposes OpenAI-compatible endpoints (/v1/chat/completions, /v1/completions, /v1/embeddings), so existing OpenAI SDK clients work with a ModelRiver base URL and API key. See OpenAI compatibility.
Can I export logs?
Yes. Use the console export tools or contact support for automated exports and SIEM integrations.
Where can I see the roadmap or changelog?
Major updates appear in the console announcements and the changelog. Ask support about joining the customer Slack channel for early insights.