Security & data handling
AVFlow is a processing plane: to mix, transcode, caption, translate, or deliver live media, the media must pass through an AVFlow Job. This page explains the resulting boundaries and the controls your team keeps.
Your application connects participants to their original RTC platform or stream endpoint. The Job then subscribes or pulls from that source, applies the requested nodes, and writes to the configured sinks. AVFlow therefore processes the media required by the Job; it does not need a custom client SDK inside your application.
Source and sink credentials are supplied in the Job JSON. Use narrowly scoped, short-lived input tokens where the platform supports them, publish-only tokens for agents that only need to speak, and least-privilege storage credentials for outputs. Never commit production credentials or paste them into a shared Job.
Recording Jobs can write to the S3-compatible destination configured in the sink. The bucket, path prefix, credentials, lifecycle policy, and public/private access policy remain yours. Public examples deliberately contain placeholders rather than usable production secrets.
A recorder only needs to subscribe; a translated-channel publisher should not subscribe to the translator it creates; an agent sink should use a publish-only token. The select filter and component inputs are the API surface for enforcing those boundaries.
Security is not only encryption and access control. For live infrastructure, the product must tell you what happened, why, and for how long.
Every Job has submitted, starting, running, stopping, completed, or failed state. Responses also expose component-level status and errors, so your backend can distinguish a bad source from a failed output.
Read the reference →Cloud webhook requests include an HMAC-SHA256 signature over the raw body. Failed deliveries retry with exponential backoff, so handlers must be idempotent by eventId.
Read the reference →Policies such as maxDurationSec and idleTimeoutSec bound how long a Job can stay alive. A deleted Job enters stopping while sinks flush and segment uploads finish.
Read the reference →The API reports cumulative component metrics and the Console usage view is authoritative for billing. AVFlow bills processed media and successful delivery, not nominal worker size.
Read the reference →Before taking an AVFlow pipeline to production, confirm these items with your own security and operations teams.
For region, retention, subprocessor, SLA, or incident-response documentation, contact support@avflow.dev. We will tell you what is documented today rather than guessing during a procurement review.