Remove model memory UI, add AGENTS.md, bump version

Add AGENTS.md with repository guidance and build/runtime notes. Remove the model memory estimation UI and related helpers: delete ModelMemoryCard, MemoryInfoRow, formatSignedBytes, and associated imports/props from AllConfigScreen.kt; remove related strings from res/values/strings.xml. Bump app versionCode to 7 and versionName to 1.3 in app/build.gradle.kts. Files changed: AGENTS.md (new), app/build.gradle.kts, app/src/main/java/.../AllConfigScreen.kt, app/src/main/res/values/strings.xml.

Author: TriDefender

AGENTS.md

@@ -0,0 +1,144 @@
+# AGENTS.md
+
+This file provides guidance to Codex (Codex.ai/code) when working with code in this repository.
+
+## Overview
+
+This is a **Jetpack Compose Android GUI wrapper** for llama.cpp. The app does NOT compile llama.cpp source code - it only packages and launches prebuilt native binaries (specifically `llama-server` and supporting libraries) on Android devices. The wrapper is designed for Snapdragon-oriented builds with CPU, OpenCL, and Hexagon/HTP acceleration support.
+
+**Architecture principle**: The Android app is a thin UI layer that manages native process lifecycle via `ProcessBuilder`. All actual inference happens in the prebuilt `llama-server` executable.
+
+## Key Architecture Components
+
+- **Native Integration**: Packages prebuilt binaries from `../pkg-snapdragon/llama.cpp/` into `app/src/main/jniLibs/arm64-v8a/`, then extracts/symlinks them to app-private storage at runtime
+- **GUI Layer**: Jetpack Compose with Material 3 and Navigation Component
+- **Dependency Injection**: Hilt (`LlamaServerApplication` with `@HiltAndroidApp`)
+- **Persistence**: DataStore for configuration storage, JSON serialization via `ConfigRepositoryImpl`
+- **Process Management**: `ServerProcessManager` launches `llama-server` as a child process and monitors stdout/stderr for status
+
+## Critical Integration Points
+
+### Binary Preparation (`util/BinaryExtractor.kt`)
+- Packages `libllama-server.so` (renamed from `bin/llama-server`) and libraries in `jniLibs/arm64-v8a/`
+- At runtime, symlinks (or copies if symlinking fails) binaries to `filesDir/bin/` and `filesDir/lib/`
+- Marks executable with `chmod 755`
+
+### Server Process Launch (`service/ServerProcessManager.kt`)
+- Builds `llama-server` command line with flags from `ServerConfig`
+- Configures environment variables for library loading:
+  - `LD_LIBRARY_PATH`: includes `filesDir`, `filesDir/lib`, and system paths
+  - `ADSP_LIBRARY_PATH`: `<filesDir>/lib`
+  - `GGML_HEXAGON_EXPERIMENTAL=1` (for HTP devices)
+- Monitors stdout/stderr for `HTTP server is listening` or `llama server listening` to mark server as running
+- **Real-time logging**: Uses `realtimeOutput` StateFlow to emit each line as read, parsed by `RuntimeViewModel` for display
+
+### Configuration Model (`model/ServerConfig.kt`)
+Comprehensive data class that maps to `llama-server` CLI flags:
+- Model selection, embedding mode, context size, batch size
+- Device types: `cpu`, `opencl`, `htp0`-`htp4`
+- KV cache types, KV offload, flash attention, auto-fit settings
+- HTTP server config (port, bind, API key), timeout, threads, continuous batching
+- `extraParams`: String for passing additional command-line arguments
+
+## Build Commands
+
+### Prerequisites
+Native artifacts must exist at:
+```text
+../pkg-snapdragon/llama.cpp/
+├── bin/llama-server
+└── lib/*.so
+```
+
+### Development Workflow
+
+1. **Copy native libraries**:
+   ```bash
+   ./copy-native-libs.sh
+   ```
+   Copies artifacts to `app/src/main/jniLibs/arm64-v8a/` and renames `llama-server` to `libllama-server.so`
+
+2. **Build with Docker** (recommended):
+   ```bash
+   docker run --rm \
+     -v /d/llama.cpp/android-gui-wrapper:/source \
+     -w /source \
+     llama-android-builder:latest \
+     bash -c "rm -rf app/build && export ANDROID_HOME=/opt/android-sdk && gradle assembleDebug --no-daemon --parallel"
+   ```
+
+3. **Build release APK** (signed):
+   ```bash
+   docker run --rm \
+     -v /d/llama.cpp/android-gui-wrapper:/source \
+     -w /source \
+     llama-android-builder:latest \
+     bash -c "rm -rf app/build && export ANDROID_HOME=/opt/android-sdk && gradle assembleRelease --no-daemon"
+   ```
+
+4. **Install on device**:
+   ```bash
+   adb install app/build/outputs/apk/debug/app-debug.apk
+   # or for release
+   adb install app/build/outputs/apk/release/app-release.apk
+   ```
+
+### Output
+- Debug APK: `app/build/outputs/apk/debug/app-debug.apk`
+- Release APK: `app/build/outputs/apk/release/app-release.apk`
+
+### Signing
+Release builds require a keystore. See `keystore.properties.example` and `README.md` (Release signing section) for setup instructions. Signing config is optional — debug builds work without it.
+
+## Runtime Flow
+
+```
+MainActivity
+└── MainScreen
+    ├── Config tab  -> AllConfigScreen (ui/launcher/)
+    └── Runtime tab -> RuntimeScreen (ui/runtime/)
+```
+
+Server startup sequence:
+```
+RuntimeScreen
+└── RuntimeViewModel.startServer()
+    └── ServerProcessManager.startServer(config)
+        ├── BinaryExtractor.ensureAllAvailable(context)
+        ├── build llama-server command line
+        ├── configure LD_LIBRARY_PATH and other env vars
+        ├── ProcessBuilder(...).start()
+        └── read stdout/stderr for server status (real-time via realtimeOutput)
+```
+
+## Common Modification Points
+
+- **Add UI settings**: Edit `AllConfigScreen.kt`, `ModelConfigViewModel.kt`, and `ServerConfig.kt`
+- **Change llama-server flags**: Modify `ServerProcessManager.kt`
+- **Modify binary/library preparation**: Update `BinaryExtractor.kt` and `copy-native-libs.sh`
+- **Change runtime controls or logs**: Edit `RuntimeScreen.kt` and `RuntimeViewModel.kt`
+- **Adjust Gradle configuration**: Modify `app/build.gradle.kts`
+
+## Important Constraints
+
+- **Only supports `arm64-v8a` ABI** - other architectures require native builds and Gradle changes
+- **Model files** use Android's document picker URI system (`UriUtils` resolves `content://` URIs)
+- **Binary compatibility**: The Android app can only use features supported by the prebuilt `llama-server` binary
+- **Package name mismatch**: Code package is `tridefender.llama.snapdragon`, but namespace/applicationId is `tridefender.llama.snapdragon`. Always use full class names in `AndroidManifest.xml` (e.g., `android:name="tridefender.llama.snapdragon.LlamaServerApplication"` instead of `android:name=".LlamaServerApplication"`)
+- **Resource imports**: Kotlin files importing R must use `import tridefender.llama.snapdragon.R`
+
+## Multi-User/Clone App Considerations
+
+On some devices (e.g., OnePlus with MultiApp feature), the app may be installed in multiple user profiles. To fully uninstall:
+```bash
+adb uninstall --user 0 tridefender.llama.snapdragon
+adb uninstall --user 999 tridefender.llama.snapdragon  # MultiApp profile
+# or
+adb shell pm uninstall --user all tridefender.llama.snapdragon
+```
+
+## Reference
+
+- Parent repo: [llama.cpp](../)
+- Server docs: [tools/server/README.md](../tools/server/README.md)
+- Build docs: [docs/build.md](../docs/build.md)

app/build.gradle.kts

@@ -16,8 +16,8 @@ android {
         applicationId = "tridefender.llama.snapdragon"
         minSdk = 24
         targetSdk = 34
-        versionCode = 6
-        versionName = "1.2"
+        versionCode = 7
+        versionName = "1.3"
 
         testInstrumentationRunner = "androidx.test.runner.AndroidJUnitRunner"
         vectorDrawables {

app/src/main/java/tridefender/llama/snapdragon/ui/launcher/AllConfigScreen.kt

@@ -31,7 +31,6 @@ import androidx.compose.ui.graphics.Color
 import androidx.compose.ui.graphics.vector.ImageVector
 import androidx.compose.ui.platform.LocalContext
 import androidx.compose.ui.res.stringResource
-import androidx.compose.ui.text.font.FontFamily
 import androidx.compose.ui.text.font.FontWeight
 import androidx.compose.ui.text.style.TextOverflow
 import androidx.compose.ui.unit.dp
@@ -41,7 +40,6 @@ import tridefender.llama.snapdragon.model.CacheType
 import tridefender.llama.snapdragon.model.DeviceType
 import tridefender.llama.snapdragon.model.FlashAttentionMode
 import tridefender.llama.snapdragon.model.MemoryFitState
-import tridefender.llama.snapdragon.model.ModelMemoryEstimate
 import tridefender.llama.snapdragon.model.PoolingType
 import tridefender.llama.snapdragon.viewmodel.ModelCatalogEntry
 import tridefender.llama.snapdragon.viewmodel.ModelConfigViewModel
@@ -121,8 +119,6 @@ fun AllConfigScreen(
         ModelSection(
             modelPath = config.modelPath,
             modelCatalog = analysis.catalog,
-            selectedEstimate = analysis.selectedEstimate,
-            freeBytes = analysis.freeBytes,
             isEmbedding = config.isEmbedding,
             poolingType = config.poolingType,
             onModelSelect = { viewModel.updateModelPath(it) },
@@ -193,8 +189,6 @@ fun AllConfigScreen(
 fun ModelSection(
     modelPath: String,
     modelCatalog: List<ModelCatalogEntry>,
-    selectedEstimate: ModelMemoryEstimate?,
-    freeBytes: Long,
     isEmbedding: Boolean,
     poolingType: PoolingType,
     onModelSelect: (String) -> Unit,
@@ -309,14 +303,6 @@ fun ModelSection(
                 Text(stringResource(R.string.browse))
             }
         }
-
-        Spacer(modifier = Modifier.height(12.dp))
-
-        ModelMemoryCard(
-            estimate = selectedEstimate,
-            freeBytes = freeBytes,
-            modifier = Modifier.fillMaxWidth()
-        )
 
         Spacer(modifier = Modifier.height(16.dp))
 
@@ -753,102 +739,6 @@ fun ExtraParamsSection(
     }
 }
 
-@Composable
-private fun ModelMemoryCard(
-    estimate: ModelMemoryEstimate?,
-    freeBytes: Long,
-    modifier: Modifier = Modifier
-) {
-    Surface(
-        modifier = modifier,
-        shape = RoundedCornerShape(12.dp),
-        color = MaterialTheme.colorScheme.surfaceVariant.copy(alpha = 0.5f)
-    ) {
-        Column(
-            modifier = Modifier
-                .fillMaxWidth()
-                .padding(16.dp),
-            verticalArrangement = Arrangement.spacedBy(10.dp)
-        ) {
-            Row(
-                modifier = Modifier.fillMaxWidth(),
-                horizontalArrangement = Arrangement.SpaceBetween,
-                verticalAlignment = Alignment.CenterVertically
-            ) {
-                Column(modifier = Modifier.weight(1f)) {
-                    Text(
-                        text = stringResource(R.string.model_memory_calculator),
-                        style = MaterialTheme.typography.bodyLarge,
-                        fontWeight = FontWeight.Medium
-                    )
-                    Text(
-                        text = stringResource(R.string.unified_memory_note),
-                        style = MaterialTheme.typography.bodySmall,
-                        color = MaterialTheme.colorScheme.onSurfaceVariant
-                    )
-                }
-
-                if (estimate != null) {
-                    AssistChip(
-                        onClick = {},
-                        label = { Text(fitLabel(estimate.fitState)) },
-                        colors = AssistChipDefaults.assistChipColors(
-                            labelColor = estimate.fitState.fitColor()
-                        )
-                    )
-                }
-            }
-
-            if (estimate == null) {
-                Text(
-                    text = stringResource(
-                        R.string.model_memory_waiting,
-                        formatBytes(freeBytes)
-                    ),
-                    style = MaterialTheme.typography.bodySmall,
-                    color = MaterialTheme.colorScheme.onSurfaceVariant
-                )
-            } else {
-                MemoryInfoRow(stringResource(R.string.available_memory), formatBytes(estimate.freeBytes))
-                MemoryInfoRow(stringResource(R.string.model_weights), formatBytes(estimate.modelBytes))
-                MemoryInfoRow(stringResource(R.string.kv_cache_estimate), formatBytes(estimate.kvCacheBytes))
-                MemoryInfoRow(stringResource(R.string.runtime_overhead), formatBytes(estimate.overheadBytes))
-                MemoryInfoRow(stringResource(R.string.total_memory_estimate), formatBytes(estimate.totalBytes))
-                MemoryInfoRow(
-                    label = stringResource(R.string.memory_after_load),
-                    value = formatSignedBytes(estimate.remainingBytes),
-                    valueColor = estimate.fitState.fitColor()
-                )
-            }
-        }
-    }
-}
-
-@Composable
-private fun MemoryInfoRow(
-    label: String,
-    value: String,
-    valueColor: Color = MaterialTheme.colorScheme.onSurface
-) {
-    Row(
-        modifier = Modifier.fillMaxWidth(),
-        horizontalArrangement = Arrangement.SpaceBetween,
-        verticalAlignment = Alignment.CenterVertically
-    ) {
-        Text(
-            text = label,
-            style = MaterialTheme.typography.bodySmall,
-            color = MaterialTheme.colorScheme.onSurfaceVariant
-        )
-        Text(
-            text = value,
-            style = MaterialTheme.typography.bodySmall,
-            fontFamily = FontFamily.Monospace,
-            color = valueColor
-        )
-    }
-}
-
 @Composable
 private fun PoolingType.localizedName(): String = when (this) {
     PoolingType.NONE -> stringResource(R.string.none)
@@ -883,11 +773,6 @@ private fun formatBytes(bytes: Long): String {
     }
 }
 
-private fun formatSignedBytes(bytes: Long): String {
-    val prefix = if (bytes > 0) "+" else ""
-    return prefix + formatBytes(bytes)
-}
-
 @Composable
 fun ConfigSectionCard(
     title: String,

app/src/main/res/values/strings.xml

@@ -170,15 +170,6 @@
     <string name="cached_models_count">%1$d cached GGUF model(s)</string>
     <string name="model_menu_memory_line">%1$s - %2$s</string>
     <string name="model_metadata_unavailable">Metadata unavailable</string>
-    <string name="model_memory_calculator">Model Memory Calculator</string>
-    <string name="unified_memory_note">Phone RAM is unified; model weights, KV cache, and runtime overhead use the same pool.</string>
-    <string name="model_memory_waiting">Select a readable GGUF model to estimate memory use. Current free memory: %1$s.</string>
-    <string name="available_memory">Available memory</string>
-    <string name="model_weights">Model weights</string>
-    <string name="kv_cache_estimate">KV cache estimate</string>
-    <string name="runtime_overhead">Runtime overhead</string>
-    <string name="total_memory_estimate">Total estimate</string>
-    <string name="memory_after_load">After load</string>
     <string name="fit_wont_fit">Won\'t fit</string>
     <string name="fit_tight_fit">Tight fit</string>
     <string name="fit_good_fit">Good fit</string>