v1.3.0

Added

Clips is by Denkiyaa, who designed, built and now maintains it — the capture pipeline, the library, the editor, the sharing step, the opt-in installer and the asm-clipd daemon (#156, #159). Bug reports and pull requests touching any of it go to them.

• Clips: save the last thirty seconds, with each Sonar channel on its own audio track. Installed from Settings rather than shipped with ASM, and deliberately so: capture is the one feature that needs packages a desktop does not already have — PyGObject, four GStreamer plugin sets and ffmpeg — and an install that only wants the mixer and the equaliser should not carry a screen recorder's encoders. Settings offers an Install button that fetches exactly those packages and nothing else, and an Uninstall that turns the feature off and offers to take them away again; removal is a separate question with No as the answer it defaults to, because ffmpeg and GStreamer belong to your video player and your browser as much as to this. If a package goes missing afterwards the button says Repair rather than claiming the feature works. A rolling capture runs in the background; the Clips page — or asm-clipd bound to a key — writes the preceding seconds to ~/Videos/ASM Clips as an .mkv carrying Game, Chat and Mic as separate tracks, so what you keep can still be remixed afterwards. The library is a grid of poster frames rather than a list of timestamps that differ only by the second they were taken; the frame is lifted from five seconds before the end, inside the span the editor opens with, because the middle of a clip is usually the part where nothing has happened yet. The editor trims, remembers the span per clip, and exports to a separate Shared/ folder so an export never comes back as another card in the library. Exports are always MP4, whatever the settings: Matroska is the only container that holds a track per channel, and it is also one Discord will upload and then refuse to play, so the recording format and the sharing format are deliberately not the same thing. Exporting ends in a share step with the poster frame as a drag-out card — drag it straight into Discord. Select, rename and delete in bulk; deletion goes through the system trash, because a mis-click on a grid of near-identical cards should be recoverable.

• Clips: the editor's preview is a mixer — every channel plays at once, with a live level and mute on each. A media player decodes exactly one audio track and picks the first, Qt's included, so a clip whose Game channel was empty played back in total silence with three perfectly good channels sitting beside it, and the only available reading was that clips record no sound. Each channel is now given a player of its own and run in step with the picture, so the sliders and Mute do what a mixer's controls do: you hear the change while you make it, and what you hear is exactly what the export will contain. Those levels are remembered per clip, alongside the trim — deciding the microphone was too loud is a judgement about that recording, and reopening a clip used to make you decide it again. Channels holding nothing are marked silent, and if every one of them is, the editor says so and points at the Sonar routing on the Home page, which is where that is actually fixed.

• Clips: the editor lost its transport bar and gained a scrubbable timeline. The play button and seek slider above the preview were a second, worse copy of the trim band below them — two scrubbers for one clip, and the one that also showed the span being exported was the one worth keeping. Click the picture to play or pause (or press Space), and drag the playhead along the band to run back and forth over a moment, which was not possible at all before: dragging inside the selection moved the selection, and everywhere else was a single click.

• Clips: choose the capture frame rate, and pin an exact one when you export. The Clips page offers 30 or 60 fps, and the editor offers As recorded, 30 or 60. These are different things and both are worth having. The screen portal only produces a frame when the picture changes, so what capture takes is a ceiling and never a promise — 30 is the new default because asking for 60 sets the keyframe interval and the encoder's budget against a rate the compositor rarely reaches. A rate can only genuinely be fixed on the way out, where the gaps can be filled, so that is where the export offers it (at the cost of a re-encode, which the dialog says up front).

• Clips: the capture rate read-out now shows both numbers. A status bar reading 20 fps that saves a 12 fps clip looks like a bug and is not one: the live figure covers the last few seconds and the file averages its whole length, idle stretches included. Both are shown when they disagree, alongside the rate the screencast itself is producing, and a save reports the rate the file actually ended up with.

• Clips: you can change what is being captured, without a terminal. The screen picker appears once and never again — the choice is persisted and replayed on every later start, which is what stops a rolling capture from prompting each time it rebuilds its pipeline. The cost was that the first answer was permanent: pick the wrong monitor, or pick a window and later want the whole screen, and the only way back was asm-clipd --forget from a terminal. There is now a Capture control beside Length and Frame rate that asks again, and if a capture is running it restarts there and then rather than springing the picker on you twenty minutes later. What is captured still cannot be displayed — the choice lives inside the portal and Wayland never tells the app what was picked — so the control offers the only honest thing, which is the way back to the picker.

• Clips: the controls were rearranged, and Select all finally has an opposite. The two things you press, Start and Save, sat in one row among the four you set once and leave alone, so they carried no more weight than a spin box; they are now a row of their own above the settings. The shortcut stopped being a loose sentence with a button after it and became a labelled control like the others. And a grid that offered Select all offered no way out of a selection: clicking empty space clears one in a file manager, but a grid of cards usually leaves no empty space to click, so the only route was ctrl-clicking every card back off.

Changed

• ASM no longer asks whether audio should go through Sonar — it just does. A dialog used to offer, once per installation, to make a Sonar channel your default output. It was asking permission for the thing ASM exists to do, seconds after launch, before you had seen a single channel; and it read the audio graph only, with no idea whether the headset was switched on, so with a wireless headset off it offered to route audio into a channel pointing at a device that was not there — which would have taken the sound away. The daemon has adopted the default on every connect since 1.2.28 anyway, with redirect_audio_on_connect in Settings to turn it off, and unlike the dialog it checks the headset is actually on first.

Fixed

• Switching Clips on could take your sound away, and keep it away. The switch opened ASM's general dependency dialog whenever a capture package was missing. That dialog lists every failing check in the app, and one of them — "pipewire-pulse running" — is fixed by restarting PipeWire and pipewire-pulse, so its Install-all button restarted the audio stack. The headset's card came back with its profile switched off, WirePlumber remembered that as a deliberate choice, and replugging the headset afterwards did not help because the remembered answer was replayed each time. Clips now installs and removes only its own packages, through its own commands, and cannot restart a service.
• Record symbols piled up in the corner of the screen, one per capture you had ever started. Three overlapping indicators, reported from a session that had stopped and restarted the capture a couple of times. They were real: each one was a live screen-share session that ASM had opened and then let go of without closing. Releasing the reference is not the same as closing the session — the portal keeps one alive until its client closes it or leaves the bus, and the app goes on running after a capture stops. So every Stop and Start, every pipeline restart, and every capture that failed to begin left another session open, each still drawing its own indicator and each still telling the desktop that the screen was being recorded. Sessions are now closed when the capture that owns them ends, including the half-made one left behind when the screen picker is cancelled.
• A clip opened outside ASM played in silence, with two perfectly loud tracks sitting inside it. Reported as "the clips have no sound — it failed to capture the audio". The capture was fine; the file was telling every player something untrue. Matroska's FlagDefault means yes when it is not written, and the muxer never writes it, so a clip carrying five channels announced all five as the default one and each player fell back to its own tie-break. Most channels in a real session hold nothing — nothing routed to Chat, the microphone muted — so what got picked was usually one of the empty ones, and the only available reading was that nothing had been recorded. Exactly one track is now flagged on the way out, chosen by how many encoded bytes each channel holds: Opus compresses silence to almost nothing, so the loudest channel is already known from the capture buffer and costs nothing to find. That beats keeping the first track, which is Game and is empty for anyone playing through the headset directly. It is a stream copy rather than a re-encode, and any failure leaves the clip exactly as it was written — an awkward default flag is worth immeasurably more than no clip.
• Clips were labelled after whatever browser happened to be playing. A clip taken while Chrome had a tab open came out named "Google Chrome", and the Clips page announced it as what was being recorded — to a user who had picked their game's window in the screen picker. Two faults met there. The list of applications a clip is never about was matched against whole names, so it caught the bare binary name chrome and missed the Google Chrome that PulseAudio actually reports; the same hole let OBS, Brave and Firefox through under their display names. And the name itself was labelled "Recording:", which it never was — it comes from the audio graph, and what is on screen was chosen in the portal picker, which Wayland never reveals to us. The list now matches on words, so display names are caught without a game called Observation being mistaken for OBS, and the label reads "Game:".
• Everything that followed the system default was filed as game audio. ASM makes one of its channels the default output when the headset connects, so a browser, a music player, system sounds — anything not routed by name — lands there. That channel was Game, not Media: the constant naming it said "media" and held Arctis_Game, because the headset's ChatMix dial reports its own two halves as "media mix" and "chat mix" and its "media" half is Game. So a podcast was equalised as game audio and then balanced against Discord by the ChatMix dial. The default is now the Media channel; the dial keeps driving Game, which is what it is for.
• The microphone EQ could be missing entirely, on an install that never opened the Sonar tab. effect_output.sonar-micro-eq is created by a filter-chain config that nothing guaranteed: the config generator covered Game, Media, Chat and Output and skipped the microphone, no package shipped the file, and its only writer was the Apply button in the mic EQ tab. Meanwhile the daemon makes that node the default input at every startup. So on an install where nobody had pressed Apply, the headset microphone was absent from the selectable inputs, and the mic input-source setting and auto-switch did nothing at all. The daemon now writes a flat config when the file is missing — never over an existing one, which would drop your mic processing.
• The Output channel's device disagreed with itself across three tabs. Setting it in the Equalizer left Channels and Settings showing the old device, and the other way round. Two causes met: the Equalizer's picker kept its own preference file that nothing else read, and the two sides did not even spell a device the same way — one saves a node.nick, the other keys its list on node.name, so a value that did travel could still fail to match. One setting is the source of truth now, both spellings resolve, and a fresh pick is held until it is saved rather than being overwritten by a stale read.
• Every diagnostic report ever filed claimed there was no settings file. The report looked one directory above where settings are actually written, so the section meant to show what you had configured — spatial audio, HRIR, the external output device, the forced quantum — always came up empty, and each of those has had to be asked for by hand in a recent issue. It reads as a clean default configuration and never was one.
• The ASM logo at the bottom of the sidebar was cut off. A seventh page pushed the column past the window. The Help icon is a little smaller now and the sidebar's own spacing a little tighter; no other icon changed.
• SteamOS: the Distrobox installer stopped part way through, looking as though it had worked. steamos-readonly disable needs root and was called without it, so it failed, the udev step returned an error and the install stopped — after printing the steps it had already completed. Reported and diagnosed by @emoeckel. (#188)
• The .deb on the releases page was missing pieces every other package had. Debian is built two ways — debian/rules for the PPA, debian/build-deb.sh for the release attachment — and the second had drifted: no arctis-stream-guard unit, and a hand-kept list of console scripts holding five of the eight entry points, so asm-stream-guard and asm-clipd were not in the package at all. A unit file and a keybinding pointed at executables that did not exist. The list is now read from pyproject.toml, and a test compares the two build paths.