Cloudflare launched Streamline as an experimental environment and open-source example for building custom video processing pipelines on its developer platform. The project focuses on cases where Cloudflare Stream’s built-in functions are insufficient, such as adding dynamic overlays to a live stream or creating a version of a hosted video with subtitles burned into the image.
Streamline distributes responsibilities among several components of the Cloudflare platform. Containers run the media engine for extended periods with predictable CPU and memory resources, while Durable Objects manage session coordination, container lifecycle, and preview routing. Workers handle control and monitoring signals and the application interface, which can be a browser application, a software agent, or an embedded system.
What does Streamline offer in practice?
The application can create a processing session, run a video pipeline, send inputs to it, monitor metrics, and then stop it. Processing continues after the controlling application disconnects or reconnects, until the application stops it or it reaches a maximum duration that prevents the session from remaining active indefinitely.
A Container hosts the media engine, which consists of a control layer written in Go and a media processor that currently uses FFmpeg for execution. The engine can receive an RTMPS stream, read hosted video through HLS from Cloudflare Stream, or receive live data from a source such as a webcam. It can also send the output to Stream Live through RTMPS or provide a low-latency preview through WebSocket.
Supported modification operations
- Applying filters such as blur and saturation.
- Adding a static image or an updatable PNG overlay during operation.
- Burning subtitles into the video.
- Adjusting encoding parameters, including the codec, bitrate, resolution, and frame rate.
Cloudflare describes use cases including modifying a live stream by adding a graphics overlay, converting hosted video into a subtitled version, and sending camera footage to a processing pipeline that can be used for previewing or video analysis. The project also includes an example application built with a Worker and an Astro interface, with functions for overlays, subtitles, filters, and picture-in-picture.
Security and session lifecycle
Cloudflare designed the model so that Stream Live input and output keys are not exposed to the control application or stored in the browser; instead, they are kept in Workers secrets or Durable Objects storage, and the application references named profiles rather than handling the keys directly. The Worker also verifies the Cloudflare Access session and associates the active session with the authorized identity, with one session allowed per deployment in the described setup.
The company cautions that the private deployment model shown is not a ready-made model for a public, multi-user service. The public preview instead uses one container identity for each authenticated user and one active session per user, along with controls for concurrency, media, and resources.
What does this mean for developers?
Streamline provides a flexible layer between managed video services and custom processing, rather than requiring developers to build the entire operational infrastructure outside the platform. However, the model’s limitations are important: current processing relies on container CPU, which may become a bottleneck when using high resolutions or frame rates. Cloudflare says future directions include computer vision pipelines, hardware-accelerated media processing, and support for WebRTC and MoQ, in addition to native encoding and decoding primitives inside Workers.
Cloudflare released two open-source repositories: cloudflare/streamline for the media engine and API packages, and cloudflare/streamline-demo for the Worker application, Astro interface, deployment tools, and Access integration. It also made a public preview available for testing or deployment within a user’s account.