@Pikacnu

Not planned, but…
AI found lots of bugs and issues.
They aren’t fixed or implemented yet (I didn’t even know that).
So… still fixing them and building the Graph System.

Still finishing graph layer to gst.
Frontend not start yet.

Finished implementing the basic function for parseNode.
Still need to update the graph to add more capabilities directly to the node system (such as removing pads from nodes, which is not yet implemented).

The node rewrite is almost complete, with only a few nodes left to migrate to the new type.
Once all tests pass and the migrations are finished, I can start working on the stream (making sure it can push and verify first using the test pipeline built on top of the node system).
After that, I’ll move on to building the frontend.

I am rewriting the node system(which allows frontend nodes to be transformed into backend nodes,and then builds a gst.pipeline to process them),
Which is still incomplete.
My goal is to let you create a pipeline with minimal media processing knowledge, while still being able to control most things and understand the process (just simpler).

Designing Nodes (GStreamer)
Wrap GStreamer components into nodes (like a box) and sync with the frontend.
But with the features GStreamer has, I need to check the differences between the frontend and backend graphs and apply them to the GStreamer pipeline.
Which is difficult, and I haven’t come up with a good idea yet.
If anyone can give me some advice, that would be great.

1. Add a simple frontend using SvelteFlow
   (This is my first time using Svelte)
2. Wrap GStreamer into node-like abstractions
   (This will sync with the frontend)

• Note: This workflow is just to get started with SvelteFlow; it’s not the final result.
• Note: The GStreamer component is still a work in progress (WIP).

Figuring out how to parse video and mux it into the target format (FLV).
Asked AI to understand how GStreamer works and how to build a proper pipeline.
Also tried to start SRT and WHIP listeners.
*WHIP doesn’t work for this use case because it can’t handle multiple audio tracks.

Try to use gst to encode / decode video (from file now)
Use 2 mpv to play video and audio(autoaudio/video won’t work and i have no idea.)
Also platfrom’s chat room fetcher (not complete yet)

There is it’s pipeline:

filesrc → matroskademux ──(video/x-av1)→ av1parse → av1dec → videoconvert → x264enc → queue → appsink ─→ mpv #1 (H.264)
                        └──(audio/x-opus)→ opusdec → queue → audioconvert → wavenc → appsink ─→ mpv #2 (WAV)

1. Write readme (helped by ai)
2. Fix bug when parsing hostname
3. add ci to build docker image

Finally I finish it
weeee

1. make function execute by sub-domain (see image below)
2. add kv store endpoint (and also dashboard)
3. make you able to make function private (still can test)
4. warp function kv_set/ delete/get into a object
   (kv.set / kv.delete / kv.get)

1. Finished the frontend (with help from AI)
2. Bug fixing (a lot of bugs… one of them was a deadlock during environment injection)
3. Passed requests to functions (it can now process requests)
4. Added more information to execution details (debug info)

Ahhh…
The frontend is almost finished,with some examples and API documentation (all in the same docs).
Built using Nuxt UI and Vue.
The next devlog should be the last one before the first release.
I’ll make it able to detect the hostname to handle function calls (similar to Cloudflare Workers).

Screenshot is part of the website

1. Add a rate limiter to API endpoints (the rate limit for each function can be modified via API).
2. Add key-value storage (accessible only in scripts).
3. Add environment variable storage (can be updated via API).
4. Change user-related APIs to admin-only access.

The basic functionality I think is needed has been completed.
Next step: build the frontend (admin panel and documentation).

1. Update schema (for user and auth.)
2. Add User and sessions.
3. Exec now can run with starlark.

(image is ai generated test code)

1. add a value driven rack mount generate
2. create a rack mount for pi slice
3. create pi slice

Img:

1. pi slice
2. slice holder
3. A small plate to adjust (fix) screw mount misplace problem

With multiple adjustment of model.
For exp: PI slice originally didn’t have hole at end of the box. It’s because without it. Power cable can’t plug in.
And multiple times adjust hole position.

Add a cli tool to update / sync config

1. Add price tip to notify price of tokens.
2. Add a json file for free models.

1. Write a simple router using the built-in net/http.
2. Design a basic database schema to store binaries and function information.
3. Create a database wrapper. (which use Transaction as main exec method)
4. Support two languages that can compile and execute from binaries.
5. Ensure both languages can generate responses (JavaScript returns an object, Starlark returns a dict).
6. Build a simple HTTP server to test functions (compile / execute / save to DB).

TODO:

1. Add support for Starlark.
2. Implement authentication.
3. Create a frontend / admin panel.

It might take some time to work on next step.
Not only because it have lots things to do, also due to I am trying to create something by 3D Printing.
Spending too much time on it…
If I can share those model. It will be nice.