Broadcast

Not yet available. Broadcast is on the roadmap. The destinations and configuration options described here reflect the intended behavior.

Broadcast lets you forward every request and response passing through the OpenDunes gateway to one or more external destinations — your observability stack, data warehouse, or LLM evaluation platform — without any extra instrumentation in your application code.

How it works#

When Broadcast is enabled on an API key or workspace, OpenDunes captures the full event envelope for each request: the model, messages, token counts, latency, cost in micro-DA, and the response. That envelope is then fanned out asynchronously to every configured destination. Your application sees no added latency.

Each destination receives a consistent payload regardless of which model or provider handled the request. You can attach a custom metadata object to your chat completion request and it will be forwarded verbatim, making it easy to correlate traces with your internal identifiers.

Supported destinations#

Configuration#

Destinations will be configurable per API key from your dashboard or via the management API. You can configure multiple destinations on a single key and each destination is toggled independently.

What gets sent#

Every broadcast event includes:

• Request timestamp, model slug, and X-Request-Id
• Input messages (roles + content) and the full response
• Token counts: prompt_tokens, completion_tokens, total_tokens
• Cost in micro-DA and the DA equivalent
• Latency in milliseconds (time-to-first-token and total)
• Any metadata you attached to the request
• The API key identifier (never the raw secret)

PII redaction governed by Algeria Law 18-07 runs before broadcast. If redaction is active on your account, the cleaned version of the payload is what destinations receive.