Keys stay server-side
Store provider credentials as managed secrets and reference them through named provider deployments.
Route supported models through a caller-authorized gateway with server-held provider keys. Enable capture to connect model-call metadata to the same service evidence Wyrd can use for independent verification.
The gateway is not itself a universal safety or verification check. It proxies model calls, enforces caller permissions, and records activity under the configured capture policy.
Store provider credentials as managed secrets and reference them through named provider deployments.
Wyrd authorizes each caller before routing the request through the configured deployment.
Enable metadata capture for caller, model, outcome, and token usage without retaining prompts or replies.
Your app sends a request to the Wyrd endpoint using its Wyrd token. The gateway resolves the deployment and signs the upstream call with the provider credential. Model replies return to the app.
Store the provider credential and create a named deployment.
Use the supported model endpoint with a name such as openai/gpt-4o.
Opt into capture, then query call metadata in Bifrost when rows become available.
Capture is intentionally opt-in. When enabled, its records can be joined to service traces and Verifier results without storing prompt or reply payloads in metadata-only mode.
A new tenant records no gateway calls until a capture policy is enabled.
Choose metadata capture with no prompt or completion payload fields.
Give teams or workloads separate named deployments and server-held provider credentials.