Repository navigation
Closed
Labels
CI/CDCI/CD procedures or container usageCI/CD procedures or container usagedocumentationDocumentation improvements or additionsDocumentation improvements or additionsencodingAudio or video encoders but not the OS interfacesAudio or video encoders but not the OS interfacesinterfaceOS input, display, or audio interfacesOS input, display, or audio interfacesperformancePerformance or latency issues, not critical but impacts usagePerformance or latency issues, not critical but impacts usagetransportUnderlying media or data transport protocolsUnderlying media or data transport protocolswebWeb components including gst-webWeb components including gst-web
Description
Will be consistently updated. Entries without assignees are all open to pull requests. Vibe coding is welcome as long as functional validation is correct.
Checkmarked entries are currently in Pull Requests. Will be unchecked if it gets closed without merging, and will be struck out and moved to the bottom if merged.
Tier 1: High Priority, blockers for v2.0.0
Tier 2: Moderate, preferred to land in v2.0.0
Tier 3: Roadmap, but not necessarily for v2.0.0
Done
-
Expose CBR for WebSockets and CRB as an option for WebRTC mode. - Assignee: @PMohanJ -
[META] Implement chunked transport for WebRTC DataChannel / Image Clipboards #74 - Assignee: @PMohanJ -
TURN-REST subproject does not conform to specifications #172 - Assignee: @ehfd -
Use one port for Data and WebSocket Streams, and embed an internal asyncio web server instead of NGINX (NGINX can be used to make things faster, optionally, but Selkies is self-contained otherwise). - Assignee: @PMohanJ -
Integrate the web components inside the Python wheel, not as a separate directory that has to be installed. - Assignee: @ehfd or @PMohanJ -
Instead of erroring with no audio server, restore server-side --disable_audio CLI option or automatically detect no PulseAudio server #147, expose https://github.com/linuxserver/pcmflux/pull/4 to Selkies - Assignee: @PMohanJ -
Bring NVENC with X11 back to working state in containers with multiple GPUs; Contact me for details, although it's likely I'll do it myself. - Assignee: @ehfd. -
Fix NVENC encoder dropping a lot of frames with Windows ChromiumD3D11VideoDecoder- Consult settings inhttps://github.com/selkies-project/selkies/blob/7dd6a93065332e5b3fe48fe5e5eb911a91f82746/src/selkies/legacy/gstwebrtc_app.py#L257-L938. The decoder initially starts frombaselineprofile and changes tohigh(open Details in [META] Road to v2.0.0 Release #227 (comment)). -
Integrate Python-Xlib code into Selkies and remove that repository (removes the ugly workaround inpyproject.toml). Abide by the LGPL license so that such code is only connected with Selkies usingimportstatements, and leave the license file in the subdirectory.(Reduce reliance on external libraries / certain upstream packages are unmaintained #85) - Assignee: @ehfd -
Coalesce fragmented environments into Selkies and incorporate externally dependent (pixelflux, pcmflux) environment variables into settings passed to dependencies (pixelflux, pcmflux) from Selkies. ChangeSELKIES_AUTO_GPU(and theAUTO_GPUfallback) so that an alternative value totrue, so e.g.,nvidia -> nvidia / amd -> amdgpu,radeon / intel -> i915,xe,gma500, etc., are mapped, and the first GPU in each vendor is chosen instead of the first GPU overall (make the input case-insensitive). Note that you should map to non-x86 GPUs as well as necessary. -
Refer to https://docs.kasm.com/docs/how-to/manual_intel_amd/index.html so that DRM KMS devices are mapped correctly in both LinuxServer.io and Selkies containers, when e.g.,/dev/dri/card0is occupied by IPMI without an additional/dev/dri/renderD*device, and the correct mapping for GPUs is instead/dev/dri/card1to/dev/dri/renderD128. Utilize the/sys/class/drm/interface, which is also available in containers. -
[Enhancement] Test client CPU and GPU consumption throughMediaStreamTrackGenerator/VideoTrackGeneratorlinuxserver/docker-baseimage-selkies#47 and Improve polyfills forMediaStreamTrackProcessorandMediaStreamTrackGenerator/VideoTrackGeneratorsupport sealskin#7 in tandem -
Deduplicate PixelFlux C and Rust code. Optimize PixelFlux C/Rust to Python frame delivery and prevent copying from C to Python; zero-copy until thewebsocketslibrary sends thememoryview. This could improve a potential AV1 implementation in the future, due to AV1 having a lot more number of packets per frame than H.264. Find a way so thathttps://github.com/linuxserver/pixelflux/blob/master/example/screen_to_browser.pydoesn't have to make a newbyteswithraw_data_from_cpp = bytes(result.data[:result.size])and also apply it to Selkies. -
Keyboard keystrokes continue to be held when packets are lost by congestion or unstable connection / Improve the last bit of keyboard latency overall while accommodating unstable connections #145 -
Most of [FEAT] The Road to WebRTC linuxserver/docker-baseimage-selkies#46 -
Reimplement congestion control mechanisms. -
Identify more zero-copy optimizations in PixelFlux. -
Compare directly inputting RGB and converting before RGB input. If usingcudaconvert, add the equivalent ofcudauploadandcudaconvertto PixelFlux so that color conversion is done by CUDA in NVENC mode as well. Default to this mode whenlibnvrtcis present and fall back to the original software color conversion when not present. -
Port and optimizehttps://github.com/selkies-project/selkies/blob/7dd6a93065332e5b3fe48fe5e5eb911a91f82746/src/selkies/legacy/gstwebrtc_app.py#L257-L938to PixelFlux. -
[META] Optimize the WebRTC stack to the maximum #157, Improve Opus: reduce NetEQ jitterbuffer & eliminate discontinuous audio artifacts from jitter with Opus + RED #153 -
Achieve higher efficiency and quality in low-bandwidth and high-latency environments #152 -
Rewire the GitHub Actions so that we now build for bothamd64andarm64using the new GitLab ARM runners. Simplify the containerization and build procedures. Integrate the web components files inside the Python wheel, not as a separate directory that has to be distributed on its own. ([META] ARM encoders, mobile/embedded devices #50) - Assignee: @ehfd -
NVIDIA Container Toolkit: Dynamically expose NVIDIA X.Org X11 display server libraries and configure the container correctly NVIDIA/nvidia-container-toolkit#563, Automatically provisioning X11 and Wayland devices of GPU inside container? NVIDIA/nvidia-container-toolkit#1477- Addressed in fix: NVIDIA X.Org driver modules mount fixes NVIDIA/nvidia-container-toolkit#1980 -
https://github.com/linuxserver/pixelflux/issues/11- Addressed in fix: Create NVIDIA control device nodes in nvidia-cdi-refresh.service NVIDIA/nvidia-container-toolkit#1979 -
Rewrite docs. -
All other possible bugs in https://github.com/selkies-project/selkies/issues?q=is%3Aissue%20state%3Aopen%20label%3Abug
Metadata
Metadata
Labels
CI/CDCI/CD procedures or container usageCI/CD procedures or container usagedocumentationDocumentation improvements or additionsDocumentation improvements or additionsencodingAudio or video encoders but not the OS interfacesAudio or video encoders but not the OS interfacesinterfaceOS input, display, or audio interfacesOS input, display, or audio interfacesperformancePerformance or latency issues, not critical but impacts usagePerformance or latency issues, not critical but impacts usagetransportUnderlying media or data transport protocolsUnderlying media or data transport protocolswebWeb components including gst-webWeb components including gst-web
Activity
ehfd commentedon Mar 27, 2026
Details for NVENC encoder dropping a lot of frames with Windows Chromium
D3D11VideoDecoder- Consult settings inhttps://github.com/selkies-project/selkies/blob/7dd6a93065332e5b3fe48fe5e5eb911a91f82746/src/selkies/legacy/gstwebrtc_app.py#L257-L938. The decoder initially starts frombaselineprofile and changes tohigh.Details
ehfd commentedon Jul 14, 2026
Example image is ready in
ghcr.io/selkies-project/selkies/example:main-ubuntu26.04.huoxingdawang commentedon Aug 6, 2026
Hi there! I was just wondering if there's any rough estimate for when v2.0.0 might be released?
ehfd commentedon Aug 6, 2026
We can get a release candidate out very soon. CI/CD has been wired up. We do have to get a couple PRs sorted out from pixelflux and pcmflux.
ztsyx commentedon Aug 13, 2026
Hi, planning to build desktop images across multiple Ubuntu bases. Could you confirm the officially supported/tested distros for the 2.0 line?
Specifically:
Context: I've hit these on 22.04:
pixelflux/pcmfluxneed GLIBC_2.34 (jammy has 2.35, focal only 2.31 — blocks 20.04?)selkies_joystick_interposer.soneeds GLIBC_2.38 (jammy 2.35 too old)libavcodec.so.60/libx264.so.164not in jammy aptQuestions:
A minimal dependency list per Ubuntu version (or a link to one) would be great. Thanks!
16 remaining items
junkerderprovinz commentedon Sep 1, 2026
v2.0.0rc0 QA, second pass
Re-ran the sweep against today's images, NVIDIA path, both bases, X11 and Wayland. All eight matrix cells passed. Five findings, line numbers are
dbc9787. Happy to expand on any of them.1. A viewer can drive player 1's gamepad.
js,is inVIEWER_ALLOWED_PREFIXES(input_handler.py:161) and the target slot istoks[2], straight from the client's message. The check that would bind a client to its own slot is gated on secure mode (selkies.py:4107,is_secure_mode = bool(master_token)at1231; same inrtc.py:1371), but the sharing fragments only apply when there is no master token, so it is off exactly where the feature lives. Connecting as?role=viewer&slot=-1and sendingjs,c,0,...gives:So a
#sharedviewer holding no slot drives player 1, againstusage.md:53("no keyboard, mouse or gamepad"). Index 99 is rejected, so the range check is the only one that fires. It also allows taking over (js,c,0,input_handler.py:4011-4021) and resetting (js,d,0,4050-4060) that pad.2. The generated certificate cannot validate anywhere it is used.
_make_self_signed_cert(stream_server.py:1207-1220) sets one SAN,DNSName("localhost"), noIPAddress, subject = issuer =CN=localhost. Stock images hit this by default:addons/base/Dockerfile:595enables HTTPS and600deletes the snakeoil pair.3. A failed start looks healthy. When
_require_configured_credentials(stream_server.py:1662-1683) exits, docker still reportsrunning restarts=0while the port returns000.addons/base/services/selkies/runis a plainexecunders6-supervisewith no restart ceiling, so it respawns forever. A ceiling or a port health check would surface it.4. Cursor callback outlives the websockets app on a mode switch.
selkies.py:5344dereferencesself.appunchecked; that attribute is set to None inshutdown()at5634, and no unregister for the callback exists (set_cursor_callbackappears twice, both registrations). Result, swallowed at5345:A race window, not a certainty. Switching to WebRTC on Wayland gave 8 occurrences on ubuntu26.04 and 7 on debiantrixie; X11 and plain WebSockets give zero.
5. The apps panel only remembers installs in the browser. State lives in
localStorageunderprootInstalledApps(selkies-web-core/lib/app-commands.js:14), both dashboards initialise from it, and there is no query command, so nothing asks the server. Another browser or a cleared cache shows installed apps as missing.Also worth noting: the "JPEG Striped Frame freezes" I mentioned earlier was my own artifact (a second browser evicting the first) and is withdrawn. Wayland is clean now, zero
swapchainand zeroNo free output buffer sloton both bases, withSELKIES_WAYLAND_X11_FALLBACK=falseand the compositor verified up first. The second screen I could not test: "Add Screen +" does nothing here and gives no feedback, but I never entered the code path, so treat it as untested rather than working.Two questions:
js,refused from viewers without a master token? I will send the PR either way.ztsyx commentedon Sep 9, 2026
Hi, after the second QA pass, do you have a rough plan for v2.0.0 stable release? Which issues are blocking stable?
ehfd commentedon Sep 9, 2026
We have the whole test suite passing, and then we get a pre-release out.
ztsyx commentedon Sep 10, 2026
Hi, is there a CI-built AppImage artifact available for testing before v2.0 is officially released?
ehfd commentedon Sep 10, 2026
@ztsyx Artifacts are in each CI job under https://github.com/selkies-project/selkies/actions/workflows/ci.yaml. Appreciate if you could give it a go.
ztsyx commentedon Sep 10, 2026
@ehfd Thanks, I'll test the AppImage from CI and report back if I find any issues.
ztsyx commentedon Sep 10, 2026
Hi @ehfd, integrated selkies-0.0.0.dev0-x86_64.AppImage from ci.yaml run 34343605118. One blocker:
After --appimage-extract, ~1992 executables under usr/conda/bin/ have shebangs pointing to /home/runner/work/selkies/selkies/AppDir/usr/conda/bin/python. Both AppRun and direct invocation fail
with exec: /home/runner/work/selkies/selkies/AppDir/usr/conda/bin/python: not found. No conda-unpack script is bundled to rewrite these on first run.
Workaround: mkdir -p /home/runner/work/selkies/selkies && ln -sfn /home/runner/work/selkies/selkies/AppDir.
ehfd commentedon Sep 10, 2026
@ztsyx Fixed in 5977f29; Please try again.
ztsyx commentedon Sep 10, 2026
@ehfd Confirmed, --appimage-extract + direct AppRun invocation both work out-of-the-box on the CI build post-5977f29. Thanks for the quick turnaround.
ehfd commentedon Sep 12, 2026
2.0.0rc0 is out. The role of this issue thread has been fulfilled.
jhony-asanuma-oxb commentedon Sep 16, 2026
hey guys, why were all 'v1.x' releases/packages deleted from
https://github.com/selkies-project/selkies/releasesandhttps://sourceforge.net/projects/selkies-gstreamer.mirror/files/?We're still using
v1.6.2, and this has broken our CI:ehfd commentedon Sep 16, 2026
@jhony-asanuma-oxb Sorry about that, but they were deleted because they were unsafe and had probably several security vulnerabilities (at least one at the very least). It's very trivial to change from v1.6.2 to v2.0.0rc0 (consider using any coding AI; it will do it), so please bear with us.
test(image): Hold item 4 to the first Ctrl+V and item 9 to a lock wit…