From c10c750c83a3cde13f7ca574dac673e2358df522 Mon Sep 17 00:00:00 2001
From: "Danilo M." <danix@danix.xyz>
Date: Sat, 19 Sep 2026 10:44:48 +0200
Subject: feat: icon-only toolbar and audio input follow-ups

Make the chat toolbar icon-only (tooltips retained) and add an icon_size config key, default 28, so the drawn glyphs can be scaled. Qt was drawing the 24px pixmaps at its ~16px default; the explicit icon size fixes that.

Fix the deferred audio minors: AudioRecorder.start() now halts a prior source and checks the source error, so a failed capture no longer claims to be recording. Send now validates every pending modality in one gate, so a mixed image+audio clip cannot reach a model without vision. Sub-second clips no longer name voice-note-0s.wav, and non-WAV audio containers classify as unknown, guarded by a test.
---
 CHANGELOG.md             |  18 ++++--
 llamachat/audio.py       |  10 +++-
 llamachat/backend.py     |   4 +-
 llamachat/config.py      |   8 +++
 llamachat/modeldialog.py |   2 +-
 llamachat/models.py      |   3 +-
 llamachat/ui.py          | 147 +++++++++++++++++++++++++++++++++--------------
 test_llamachat.py        |   5 ++
 8 files changed, 144 insertions(+), 53 deletions(-)

diff --git a/CHANGELOG.md b/CHANGELOG.md
index a2a3635..c685d54 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -69,10 +69,10 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
   moment once used. The text comes from the source markdown, since Qt emits
   one `<pre>` per line and the block cannot be recovered from the rendered
   HTML.
-- Icons on the chat buttons (New, Attach, Clear files, Stop, Send), drawn
-  at runtime so there is still no icon file to install. Send is now a square
-  button beside the entry box with its label under the icon, and New/Attach
-  sit at the right end of the picker row.
+- Icons on the chat buttons (New, Attach, Clear files, Stop, Send, Record),
+  drawn at runtime so there is still no icon file to install. The toolbar
+  buttons are icon-only, each with a tooltip, and `icon_size` in config.toml
+  sets the glyph size.
 - Skills. Instruction files read from `~/.agents/skills/*/SKILL.md` can be
   loaded into the model's context on demand: the model calls a `load_skill`
   tool, or you type `/skill-name` or mention a skill's name. Loaded skills
@@ -118,6 +118,16 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
   assistant tool-call message. Interleaved-thinking models (DeepSeek V3.2+/V4,
   GLM-4.7+ on SiliconFlow) require this and stop answering — ending the turn at
   the thinking — when it is dropped.
+- Audio capture no longer claims to be recording when it is not.
+  `AudioRecorder.start()` halts any previous source, checks the new source's
+  error, and reports a generic failure; a second start without a stop no
+  longer leaks the first source.
+- A pending image attached before a model switch could reach a model without
+  vision when sent alongside audio. The send-time gate now checks every
+  pending modality at once and offers a single switch to a model that accepts
+  them all.
+- A sub-second recording could be named `voice-note-0s.wav`; the history
+  marker now rounds to a minimum of one second.
 
 Tool calling and reasoning output vary between providers. Web search on a
 cloud model may not work as reliably as it does with a local router.
diff --git a/llamachat/audio.py b/llamachat/audio.py
index 67b568c..58c7dd6 100644
--- a/llamachat/audio.py
+++ b/llamachat/audio.py
@@ -80,10 +80,13 @@ if AVAILABLE:
             self._timer.timeout.connect(self._on_cap)
 
         def start(self) -> bool:
-            """Begin capture. False when there is no input device."""
+            """Begin capture. False when there is no input device or it fails."""
             device = QMediaDevices.defaultAudioInput()
             if device.isNull():
                 return False
+            # A second start() without a stop() would leak the previous source
+            # and leave two captures feeding two timers.
+            self._halt()
             fmt = QAudioFormat()
             fmt.setSampleRate(RATE)
             fmt.setChannelCount(CHANNELS)
@@ -93,7 +96,10 @@ if AVAILABLE:
             self._io = QBuffer(self._buffer)
             self._io.open(QIODevice.WriteOnly)
             self._source = QAudioSource(device, fmt, self)
-            self._source.start(self._io)
+            error = self._source.start(self._io)
+            if error is not None:
+                self._halt()
+                return False
             self._timer.start(MAX_SECONDS * 1000)
             return True
 
diff --git a/llamachat/backend.py b/llamachat/backend.py
index 8b9115c..cfdecde 100644
--- a/llamachat/backend.py
+++ b/llamachat/backend.py
@@ -104,7 +104,7 @@ class SkillsConfig:
 class Attachment:
     """A file the user attached, ready for both the API and the database."""
     path: Path
-    kind: str            # 'text' | 'image'
+    kind: str            # 'text' | 'image' | 'audio'
     mime: str
     size: int
     sha256: str
@@ -188,7 +188,7 @@ def build_user_content(text: str, attachments: list[Attachment]):
     """Assemble one OpenAI-format user message from text plus attachments.
 
     Returns a plain string when there is nothing but text, and the
-    multi-part content array when images are involved.
+    multi-part content array when images or audio are involved.
     """
     parts: list[str] = []
     for att in attachments:
diff --git a/llamachat/config.py b/llamachat/config.py
index f7dab5d..cfcda57 100644
--- a/llamachat/config.py
+++ b/llamachat/config.py
@@ -36,6 +36,9 @@ DEFAULTS = {
     "attach_ctx_fraction": 0.5,
     # Rough chars-per-token used to turn a ctx-size into a char budget.
     "chars_per_token": 3.5,
+    # Toolbar button glyph size in pixels. The icons are drawn, not loaded,
+    # so this is the only knob for their on-screen size.
+    "icon_size": 28,
     # Named prompt selected for new conversations. Empty means the global
     # default.md, and "none" means no system prompt at all.
     "default_prompt": "",
@@ -86,6 +89,7 @@ class Config:
     request_timeout: int
     attach_ctx_fraction: float
     chars_per_token: float
+    icon_size: int
     default_prompt: str
     prompts_dir: Path
     state_path: Path
@@ -157,6 +161,7 @@ def load(path: Path = CONFIG_PATH) -> Config:
         request_timeout=int(values["request_timeout"]),
         attach_ctx_fraction=float(values["attach_ctx_fraction"]),
         chars_per_token=float(values["chars_per_token"]),
+        icon_size=int(values["icon_size"]),
         default_prompt=str(values["default_prompt"]),
         prompts_dir=path.parent / "prompts",
         # Window layout, remembered between runs. Not user-editable config,
@@ -211,6 +216,9 @@ def write_default(path: Path = CONFIG_PATH) -> Path:
         f'attach_ctx_fraction = {DEFAULTS["attach_ctx_fraction"]}\n'
         f'chars_per_token = {DEFAULTS["chars_per_token"]}\n'
         '\n'
+        '# Toolbar button icon size in pixels.\n'
+        f'icon_size = {DEFAULTS["icon_size"]}\n'
+        '\n'
         '# Web search through SearXNG. The model decides when to search; it\n'
         '# is offered a web_search tool and calls it when a question needs\n'
         '# current information. Both search_enabled and search_url are\n'
diff --git a/llamachat/modeldialog.py b/llamachat/modeldialog.py
index c6f7501..a29a1d8 100644
--- a/llamachat/modeldialog.py
+++ b/llamachat/modeldialog.py
@@ -99,7 +99,7 @@ def to_fields(info: ModelInfo) -> tuple[str, bool, str, str, bool]:
 
 
 class ModelDialog(QDialog):
-    """Context size, vision and prices for one model.
+    """Context size, vision, audio and prices for one model.
 
     Prefilled from the provider's defaults, so the common case is checking
     the numbers rather than typing them.
diff --git a/llamachat/models.py b/llamachat/models.py
index e2f7922..fb9d9fd 100644
--- a/llamachat/models.py
+++ b/llamachat/models.py
@@ -170,7 +170,8 @@ class ModelStore:
                 section.pop(key, None)
             elif isinstance(value, bool):
                 # Before the numeric branch: bool is a subclass of int, so
-                # str(False) would otherwise write "False" for vision.
+                # str(False) would otherwise write "False" for vision and
+                # audio.
                 section[key] = "true" if value else "false"
             else:
                 section[key] = str(value)
diff --git a/llamachat/ui.py b/llamachat/ui.py
index 91321e4..b45daad 100644
--- a/llamachat/ui.py
+++ b/llamachat/ui.py
@@ -22,7 +22,8 @@ import re
 from pathlib import Path
 
 from PySide6.QtCore import (
-    QObject, QPointF, QRect, QRectF, QSettings, Qt, QThread, QTimer, Signal, Slot,
+    QObject, QPointF, QRect, QRectF, QSettings, QSize, Qt, QThread, QTimer,
+    Signal, Slot,
 )
 from PySide6.QtGui import (
     QAction, QColor, QDesktopServices, QIcon, QKeySequence, QPainter,
@@ -58,12 +59,17 @@ COPY_SCHEME = "x-llamachat-copy:"
 # The buttons get their glyphs drawn at runtime, so there is no icon file
 # to install. All the marks share the brand blue, which also makes each
 # button an unambiguous control rather than a bare text label.
-def _button_icon(kind: str) -> QIcon:
-    """A small monochrome glyph for one of the chat buttons."""
-    pixmap = QPixmap(24, 24)
+def _button_icon(kind: str, size: int = 24) -> QIcon:
+    """A small monochrome glyph for one of the chat buttons.
+
+    The glyph coordinates below are authored on a 24x24 grid and the painter
+    is scaled, so one call renders crisply at any configured icon size.
+    """
+    pixmap = QPixmap(size, size)
     pixmap.fill(QColor(0, 0, 0, 0))
     painter = QPainter(pixmap)
     painter.setRenderHint(QPainter.Antialiasing)
+    painter.scale(size / 24.0, size / 24.0)
     colour = QColor("#3F5EFB")
     pen = QPen(colour, 2.2)
     pen.setCapStyle(Qt.RoundCap)
@@ -124,6 +130,16 @@ def _button_icon(kind: str) -> QIcon:
     return QIcon(pixmap)
 
 
+def _set_button_icon(button, kind: str, size: int) -> None:
+    """Give a toolbar button its glyph and match the icon box to it.
+
+    Without the explicit icon size, Qt draws the pixmap at the style's small
+    default (about 16px) regardless of how large it was rendered.
+    """
+    button.setIcon(_button_icon(kind, size))
+    button.setIconSize(QSize(size, size))
+
+
 class ContextMeter(QWidget):
     """A bar showing how much of the model's context the next request uses.
 
@@ -694,10 +710,10 @@ class ChatWindow(QMainWindow):
         compose.addWidget(self.input, 1)
 
         # A square send control beside the entry box, like a chat app.
+        size = self.cfg.icon_size
         self.send_button = QToolButton()
-        self.send_button.setIcon(_button_icon("send"))
-        self.send_button.setText("Send")
-        self.send_button.setToolButtonStyle(Qt.ToolButtonTextUnderIcon)
+        _set_button_icon(self.send_button, "send", size)
+        self.send_button.setToolButtonStyle(Qt.ToolButtonIconOnly)
         self.send_button.setFixedSize(44, 44)
         self.send_button.setToolTip("Send (Ctrl+Enter)")
         self.send_button.clicked.connect(self.send)
@@ -750,33 +766,36 @@ class ChatWindow(QMainWindow):
         buttons.addWidget(self.prompt_button)
 
         buttons.addStretch()
-        self.clear_attach_button = QPushButton("Clear files")
-        self.clear_attach_button.setIcon(_button_icon("clear"))
+        self.clear_attach_button = QPushButton()
+        _set_button_icon(self.clear_attach_button, "clear", size)
+        self.clear_attach_button.setToolTip("Clear attached files")
         self.clear_attach_button.clicked.connect(self.clear_attachments)
         self.clear_attach_button.hide()
         buttons.addWidget(self.clear_attach_button)
 
-        self.stop_button = QPushButton("Stop")
-        self.stop_button.setIcon(_button_icon("stop"))
+        self.stop_button = QPushButton()
+        _set_button_icon(self.stop_button, "stop", size)
+        self.stop_button.setToolTip("Stop generating")
         self.stop_button.clicked.connect(self.stop_stream)
         self.stop_button.hide()
         buttons.addWidget(self.stop_button)
 
-        self.new_button = QPushButton("New")
-        self.new_button.setIcon(_button_icon("new"))
+        self.new_button = QPushButton()
+        _set_button_icon(self.new_button, "new", size)
         self.new_button.setToolTip("Start a fresh conversation (Ctrl+N)")
         self.new_button.clicked.connect(self.new_session)
         buttons.addWidget(self.new_button)
 
-        self.record_button = QPushButton("Record")
-        self.record_button.setIcon(_button_icon("record"))
+        self.record_button = QPushButton()
+        _set_button_icon(self.record_button, "record", size)
         self.record_button.setToolTip("Record audio for an audio-capable model")
         self.record_button.clicked.connect(self._toggle_record)
         self.record_button.hide()
         buttons.addWidget(self.record_button)
 
-        self.attach_button = QPushButton("Attach…")
-        self.attach_button.setIcon(_button_icon("attach"))
+        self.attach_button = QPushButton()
+        _set_button_icon(self.attach_button, "attach", size)
+        self.attach_button.setToolTip("Attach files")
         self.attach_button.clicked.connect(self._on_attach_clicked)
         buttons.addWidget(self.attach_button)
 
@@ -994,39 +1013,73 @@ class ChatWindow(QMainWindow):
                 names.append(name)
         return names
 
-    def audio_models(self) -> list[str]:
+    def _modality_models(self, audio: bool, vision: bool) -> list[str]:
+        """Model ids that satisfy every required modality.
+
+        Audio is strict, vision lenient: only a model that reports audio can
+        receive a recording, while an unknown vision status gets the benefit
+        of the doubt. Same asymmetry as the attach-time gate.
+        """
         names = []
         for i in range(self.model_box.count()):
             name = self.model_box.itemData(i)
-            if self.model_info(name).audio is True:
-                names.append(name)
+            info = self.model_info(name)
+            if audio and info.audio is not True:
+                continue
+            if vision and info.vision is False:
+                continue
+            names.append(name)
         return names
 
-    def _ensure_audio_model(self) -> bool:
-        """Offer to switch to an audio-capable model. True when one is active.
+    def _ensure_modalities(self, *, audio: bool, vision: bool) -> bool:
+        """Offer a single switch to a model that accepts everything pending.
 
-        Unlike vision, an unknown model is not given the benefit of the
-        doubt: audio input is rare and explicitly reported, so only a model
-        known to accept it is allowed to receive a recording.
+        Returns True when the current model can receive the attachments. One
+        combined switch, rather than a vision switch then an audio switch,
+        stops a mixed clip ping-ponging between a vision-only and an
+        audio-only model.
         """
-        if self.current_info().audio is True:
+        info = self.current_info()
+        if (not audio or info.audio is True) and (
+            not vision or info.vision is not False
+        ):
             return True
 
-        candidates = self.audio_models()
+        candidates = self._modality_models(audio, vision)
         if not candidates:
-            QMessageBox.warning(
-                self,
-                "No audio model",
-                "No model on the router reports audio input, so a recording "
-                "cannot be sent.",
-            )
+            if audio and vision:
+                title = "No compatible model"
+                body = (
+                    "No model accepts both audio and images, so this clip "
+                    "cannot be sent."
+                )
+            elif audio:
+                title = "No audio model"
+                body = (
+                    "No model on the router reports audio input, so a "
+                    "recording cannot be sent."
+                )
+            else:
+                title = "No vision model"
+                body = (
+                    "No model on the router has an mmproj file configured, so "
+                    "images cannot be sent."
+                )
+            QMessageBox.warning(self, title, body)
             return False
 
+        if audio and vision:
+            need = "Audio and images need a model that accepts both."
+        elif audio:
+            need = "Audio needs a model that accepts voice input."
+        else:
+            need = "Images need a vision-capable model."
+
         target = candidates[0]
         answer = QMessageBox.question(
             self,
             "Switch model?",
-            f"Audio needs a model that accepts voice input.\n\n"
+            f"{need}\n\n"
             f"Switch to {target}?\n\n"
             "The router unloads the current model to do this, so the next "
             "reply will take several seconds to start.",
@@ -1267,9 +1320,15 @@ class ChatWindow(QMainWindow):
             self.recorder is not None and self.current_info().audio is True
         )
         self.record_button.setVisible(can)
-        self.record_button.setText("Stop" if self.recording else "Record")
-        self.record_button.setIcon(
-            _button_icon("stop" if self.recording else "record")
+        _set_button_icon(
+            self.record_button,
+            "stop" if self.recording else "record",
+            self.cfg.icon_size,
+        )
+        self.record_button.setToolTip(
+            "Stop recording"
+            if self.recording
+            else "Record audio for an audio-capable model"
         )
 
     def _toggle_record(self) -> None:
@@ -1280,7 +1339,7 @@ class ChatWindow(QMainWindow):
         if self.recorder is None or self.current_info().audio is not True:
             return
         if not self.recorder.start():
-            self.show_status("No microphone available.", error=True)
+            self.show_status("Could not start recording.", error=True)
             return
         self.recording = True
         self.show_status("Recording. Click Stop when done.")
@@ -1306,7 +1365,7 @@ class ChatWindow(QMainWindow):
         else:
             self.hide_status()
         self.attachments.append(
-            backend.audio_attachment(audio.to_wav(pcm), round(seconds))
+            backend.audio_attachment(audio.to_wav(pcm), max(1, round(seconds)))
         )
         self.update_attach_label()
 
@@ -1346,12 +1405,14 @@ class ChatWindow(QMainWindow):
         if not text and not self.attachments:
             return
 
-        if any(a.kind == "audio" for a in self.attachments):
+        needs_audio = any(a.kind == "audio" for a in self.attachments)
+        needs_vision = any(a.kind == "image" for a in self.attachments)
+        if needs_audio or needs_vision:
             # A switch here changes the model, so this must run before the
-            # model is read below.
-            if not self._ensure_audio_model():
+            # model is read below. One combined gate covers a mixed clip.
+            if not self._ensure_modalities(audio=needs_audio, vision=needs_vision):
                 return
-            if not text:
+            if needs_audio and not text:
                 text = backend.AUDIO_PROMPT
 
         model = self.current_model()
diff --git a/test_llamachat.py b/test_llamachat.py
index bf43c77..f665ee3 100755
--- a/test_llamachat.py
+++ b/test_llamachat.py
@@ -251,6 +251,11 @@ def test_classify():
     assert backend.classify(Path("a.so")) == "unknown"
     assert backend.classify(Path("a.wav")) == "audio"
     assert backend.classify(Path("a.WAV")) == "audio"
+    # Only WAV is accepted; other audio containers must not be labelled
+    # audio, since the request hardcodes format "wav".
+    assert backend.classify(Path("a.mp3")) == "unknown"
+    assert backend.classify(Path("a.ogg")) == "unknown"
+    assert backend.classify(Path("a.flac")) == "unknown"
     print("ok  file classification")
 
 
-- 
cgit v1.2.3