neoagent 3.1.1-beta.9 → 3.2.0

package/.env.example

@@ -236,8 +236,8 @@ TRELLO_API_KEY=your-trello-api-key
 BRAVE_SEARCH_API_KEY=your-brave-search-api-key-here
 
 # Deepgram API key — used for:
-#   • Recordings transcription (Deepgram Nova-3 multilingual via /api/recordings)
-# Without this key: recordings can still be captured and stored, but transcription will fail until retried with a valid key.
+#   • Voice transcription (Deepgram Nova-3 multilingual)
+# Without this key: Deepgram-backed voice transcription is unavailable.
 # Get your key at: https://console.deepgram.com/
 DEEPGRAM_API_KEY=your-deepgram-api-key-here
 DEEPGRAM_BASE_URL=https://api.deepgram.com

package/README.md

@@ -1,5 +1,5 @@
 <p align="center">
-  <img src="flutter_app/assets/branding/app_icon_128.png" width="80" alt="NeoAgent">
+  <img src="marketing/final/github-banner.png" alt="NeoAgent — an AI assistant that keeps working" width="100%">
 </p>
 
 <h1 align="center">NeoAgent</h1>
@@ -28,6 +28,10 @@
   <a href="https://www.producthunt.com/products/neoagent-2?embed=true&utm_source=badge-featured&utm_medium=badge&utm_campaign=badge-neoagent-2" target="_blank" rel="noopener noreferrer"><img alt="NeoAgent - The next-gen self-hosted AI agent that works beyond chat | Product Hunt" width="250" height="54" src="https://api.producthunt.com/widgets/embed-image/v1/featured.svg?post_id=1171573&theme=light&t=1781447653039"></a>
 </p>
 
+<p align="center">
+  <img src="marketing/final/trailer.gif" alt="NeoAgent trailer" width="100%">
+</p>
+
 <p align="center">
   <img src="demo.gif" alt="NeoAgent demo" width="100%">
 </p>

package/docs/architecture.md

@@ -36,7 +36,7 @@ bridge files and native Android code where required.
 
 NeoAgent uses one `better-sqlite3` database instance. It stores users, agents,
 settings, conversations, runs, tasks, integration connections, memory,
-recordings, health data, permissions, and operational state.
+health data, permissions, and operational state.
 
 Runtime files live outside the package source under `NEOAGENT_HOME`. Schema
 changes are applied through the migration layer rather than ad hoc service SQL.

package/docs/benchmarking.md

@@ -1,102 +1,83 @@
 # Benchmarking
 
-NeoAgent includes an end-to-end benchmark harness that drives the app through
-its authenticated HTTP surface instead of calling internal engine methods
-directly. The harness configures OpenRouter models, submits benchmark tasks,
-collects run events and usage data, and writes normalized result artifacts.
+NeoAgent's benchmark harness measures the actual quality of its memory system on
+**LoCoMo** ([Snap Research](https://github.com/snap-research/locomo), CC BY-NC 4.0), the
+standard long-term conversational memory benchmark. It follows the same protocol
+mem0/Zep/Letta/[Omi](https://github.com/BasedHardware/omi-memory-benchmarks) use to publish
+their numbers, so results are comparable: a thin QA head over the memory system's real
+retrieval API, not the consumer chat agent.
 
 ## Latest dashboard
 
 ![Latest benchmark dashboard](/benchmarks/latest-dashboard.png)
 
-The generated image is backed by these local outputs:
+Backed by these local outputs:
 
-- `benchmark/results/latest-results.json`
-- `benchmark/results/latest-summary.json`
-- `benchmark/results/latest-summary.md`
+- `benchmark/results/latest-results.json` — every question, gold answer, generated
+  answer, and CORRECT/WRONG verdict.
+- `benchmark/results/latest-summary.json` / `latest-summary.md` — overall and
+  per-category accuracy.
 - `static/benchmarks/latest-dashboard.png`
 
 ## Commands
 
 ```bash
-npm run benchmark:setup
-npm run benchmark:run
-npm run benchmark:report
+npm run benchmark:setup   # downloads and validates the LoCoMo dataset
+npm run benchmark:run     # ingests, retrieves, answers, judges, scores
+npm run benchmark:report  # re-renders the dashboard from the last run's results.json
 ```
 
-## Model selection
-
-The benchmark harness is OpenRouter-first. Edit the `OPENROUTER_MODEL_IDS`
-variable in `benchmark/config.js` to pin exact model IDs, or leave it empty to
-use the current cheap-tier OpenRouter models discovered at runtime.
-
-## Benchmark coverage
-
-- `GAIA`, `BrowseComp`, `WebArena`, `VisualWebArena`, and `SWE-bench` are wired
-  as exact public-suite adapters. When their official datasets or external
-  runners are not installed, they are reported as blocked instead of being
-  approximated.
-- `NeoAgent Representative Tasks` runs first-party agent tasks through
-  `/api/agents` and scores them from the persisted run details.
-- `NeoAgent Memory Retrieval` seeds memories over `/api/memory`, recalls them
-  through the app routes, and scores retrieval quality with the same metric
-  code used by the server evaluation module.
-
-## Reproducibility
-
-- Public-suite setup manifests are written under `benchmark/workdir/<suite>/`.
-- The harness uses one benchmark account and persists per-run usage summaries
-  through the existing NeoAgent run detail endpoints.
-- Blocked suites are intentional and explicit. The harness never substitutes a
-  simplified local task and labels it as a public benchmark result.
-
-## Public benchmark prerequisites
-
-### GAIA
-
-- Access to the gated [GAIA dataset](https://huggingface.co/datasets/gaia-benchmark/GAIA).
-- `HF_TOKEN` in the environment.
-- A Python environment with `datasets` installed if you want `benchmark:setup`
-  to export a normalized local cache automatically.
-- Either an exact evaluator wired through
-  `NEOAGENT_BENCHMARK_GAIA_RUNNER` or a repository-local exact runner added to
-  the harness. The current adapter intentionally reports `blocked` until an
-  official evaluator path is configured.
-
-### BrowseComp
-
-- A checkout of [openai/simple-evals](https://github.com/openai/simple-evals),
-  which `benchmark:setup` can clone into `benchmark/workdir/browsecomp/`.
-- Exact BrowseComp case extraction from that repo.
-- An official evaluator command exposed through
-  `NEOAGENT_BENCHMARK_BROWSECOMP_RUNNER`.
-
-### WebArena
-
-- Docker installed and running.
-- The official [WebArena](https://webarena.dev/) environment and task sites.
-- Browser automation dependencies required by the official runner.
-- An exact runner command exposed through
-  `NEOAGENT_BENCHMARK_WEBARENA_RUNNER`.
-
-### VisualWebArena
-
-- Docker installed and running.
-- The official VisualWebArena environment.
-- A vision-capable configured benchmark model.
-- An exact runner command exposed through
-  `NEOAGENT_BENCHMARK_VISUAL_WEBARENA_RUNNER`.
-
-### SWE-bench
-
-- Docker installed and running.
-- A checkout of [SWE-bench](https://github.com/SWE-bench/SWE-bench), which
-  `benchmark:setup` can clone into `benchmark/workdir/swebench/`.
-- The official dataset/runtime and enough disk/network budget for repo/image
-  setup.
-- An exact runner command exposed through
-  `NEOAGENT_BENCHMARK_SWEBENCH_RUNNER`.
-
-On this machine, `docker` is currently missing, so exact `WebArena`,
-`VisualWebArena`, and `SWE-bench` execution remain blocked until Docker and the
-official runners are installed.
+`benchmark:run` requires:
+- A running NeoAgent server (`NEOAGENT_BENCHMARK_BASE_URL`, default `http://127.0.0.1:3333`).
+- `OPENROUTER_API_KEY` in the environment — it configures each benchmark account's own
+  memory-extraction provider, and drives the answerer/judge models.
+
+## How it works
+
+1. **Ingestion.** For each LoCoMo conversation, the harness registers an isolated
+   benchmark account and pushes every session through NeoAgent's real document-ingestion
+   pipeline (`POST /api/memory/ingestion/documents`) — one document per session, dated and
+   speaker-labeled, chunked exactly like any other real ingested source.
+2. **Extraction draining.** NeoAgent promotes raw ingested chunks into structured
+   memories/facts as a side effect of the agent loop's memory-recall step, not the raw
+   ingestion call itself. The harness issues a handful of trivial agent runs per account to
+   settle this before measuring retrieval — using the app's real settling mechanism rather
+   than reaching into internals.
+3. **Retrieval.** `POST /api/memory/memories/recall` — NeoAgent's real fused
+   lexical+entity+vector retrieval surface.
+4. **Answering.** One OpenRouter chat completion per question, using mem0's verbatim
+   `ANSWER_PROMPT` (≤5-6-word answers, relative→absolute date conversion).
+5. **Judging.** One OpenRouter chat completion using mem0's verbatim `ACCURACY_PROMPT`
+   (lenient CORRECT/WRONG) — the same judge all of the above vendors report against.
+   LoCoMo category 5 (adversarial) is excluded, exactly as mem0's own eval code does.
+
+Prompts live verbatim in [`benchmark/locomo/prompts.js`](../benchmark/locomo/prompts.js).
+
+## Configuration
+
+| Env var | Default | Purpose |
+| --- | --- | --- |
+| `NEOAGENT_BENCHMARK_BASE_URL` | `http://127.0.0.1:3333` | Server to benchmark |
+| `OPENROUTER_API_KEY` | — (required) | Answerer/judge models + per-account memory extraction |
+| `BENCHMARK_ANSWER_MODEL` / `BENCHMARK_JUDGE_MODEL` | cheapest priced OpenRouter model | Pin exact model IDs |
+| `BENCHMARK_LOCOMO_SAMPLES` / `BENCHMARK_LOCOMO_OFFSET` | `10` / `0` | Which conversations to run |
+| `BENCHMARK_LOCOMO_QA_PER_SAMPLE` | `0` (all) | Cap questions per conversation |
+| `BENCHMARK_LOCOMO_MEM_K` | `20` | Recall depth per question |
+| `BENCHMARK_CONCURRENCY` | `1` | Parallel QA-head requests |
+| `BENCHMARK_CASE_LIMIT` | `0` (no cap) | Global question cap, useful for smoke tests |
+
+For a cheap smoke test: `BENCHMARK_LOCOMO_SAMPLES=1 BENCHMARK_LOCOMO_QA_PER_SAMPLE=5 npm run benchmark:run`.
+
+## Honest caveats
+
+- **Single run, temperature 0.** No averaging across repeats.
+- **Category 5 (adversarial) excluded**, matching the standard mem0 LoCoMo protocol.
+- **Ingestion granularity.** LoCoMo sessions are ingested as NeoAgent's native document
+  unit (one per session); this differs from message-pair-level ingestion some vendors use,
+  which can make retrieval comparatively easier or harder depending on the question.
+- **Judge leniency** varies significantly across published papers; this harness uses mem0's
+  lenient judge verbatim so numbers line up with mem0/Zep/Letta/Omi's published columns.
+- The LoCoMo dataset (CC BY-NC 4.0) is downloaded on demand into `benchmark/workdir/`
+  (gitignored) — never vendored into the repository. Result files may quote benchmark
+  questions/answers for verifiability; no NeoAgent user data is involved (every run uses
+  synthetic benchmark accounts).

package/docs/capabilities.md

@@ -14,7 +14,7 @@ capabilities by workflow instead of maintaining one long inventory.
 
 - [Integrations and messaging](integrations.md)
 - [Devices and interfaces](devices.md)
-- [Recordings and health](recordings-and-health.md)
+- [Health](health.md)
 
 ## Run unattended work
 

package/docs/clients-and-devices.md

@@ -21,10 +21,9 @@ updates into `MainController`, which notifies the relevant UI.
 
 ## Android bridges
 
-Native Kotlin code handles Health Connect, background recording,
-notifications, launcher functions, telecom integration, and Android home
-widgets. Flutter bridges normalize permission and lifecycle behavior before
-updating application state.
+Native Kotlin code handles Health Connect, notifications, launcher functions,
+telecom integration, and Android home widgets. Flutter bridges normalize
+permission and lifecycle behavior before updating application state.
 
 ## Desktop and Chrome pairing
 

package/docs/configuration.md

@@ -49,7 +49,7 @@ neoagent env set SESSION_SECRET "$(openssl rand -hex 32)"
 | `ANTHROPIC_BASE_URL` | Anthropic-compatible base URL override |
 | `OLLAMA_URL` | Ollama server URL |
 | `BRAVE_SEARCH_API_KEY` | Web search |
-| `DEEPGRAM_API_KEY` | Recording transcription |
+| `DEEPGRAM_API_KEY` | Voice transcription |
 | `DEEPGRAM_BASE_URL` | Deepgram-compatible base URL override |
 | `DEEPGRAM_MODEL` | Speech model |
 | `DEEPGRAM_LANGUAGE` | Speech language mode |

package/docs/devices.md

@@ -7,7 +7,7 @@ explicitly paired as a runtime backend.
 ## Web interface
 
 The web interface provides chat, runs, tasks, agents, integrations, memory,
-devices, recordings, health, settings, permissions, and logs. It is served by
+devices, health, settings, permissions, and logs. It is served by
 the NeoAgent backend.
 
 ## Android app
@@ -15,7 +15,6 @@ the NeoAgent backend.
 The Android app connects to the same backend and adds:
 
 - Health Connect synchronization
-- Background microphone recording
 - Android notification forwarding
 - Native notifications and device pairing
 

package/docs/health.md

@@ -0,0 +1,24 @@
+# Health
+
+Health data is an optional feature of the NeoAgent clients. It stores
+processed data on the NeoAgent backend.
+
+## Health Connect
+
+The Android app can read Health Connect records after the user grants Android
+permissions. Supported data includes steps, heart rate, sleep, exercise, and
+weight.
+
+The app uploads granted records to the user's NeoAgent backend. Stored metrics
+appear in **Health** and are available to the owning agent through the health
+tool.
+
+Health Connect access can be revoked in Android settings. Revoking access stops
+future reads but does not remove records already uploaded to NeoAgent; delete
+stored server data separately when required.
+
+## Privacy
+
+Health records are sensitive personal data. Protect the server account, use
+HTTPS for remote clients, restrict agents and integrations, and include this
+data store in backup and deletion procedures.

package/docs/index.md

@@ -8,7 +8,7 @@ sidebar_label: Overview
 
 NeoAgent is a self-hosted AI agent that runs continuously on a macOS or Linux
 server. It provides an operator interface for chat, automation, integrations,
-memory, recordings, and connected devices.
+memory, and connected devices.
 
 NeoAgent is beta software. Install it on a machine you administer, start with
 restricted tool permissions, and read the security guide before connecting
@@ -42,7 +42,7 @@ Continue with:
 | [Memory](memory.md) | What NeoAgent remembers and how to manage it |
 | [Integrations](integrations.md) | Connected app accounts and messaging channels |
 | [Devices and interfaces](devices.md) | Android, desktop, Chrome, launcher, and wearable surfaces |
-| [Recordings and health](recordings-and-health.md) | Audio transcription and Health Connect |
+| [Health](health.md) | Health Connect |
 | [Skills and MCP](skills.md) | Reusable instructions and external tool servers |
 | [Operations](operations.md) | Updates, backups, logs, and recovery |
 | [Billing](billing.md) | Stripe subscriptions, plans, trials, and webhooks |

package/docs/integrations-architecture.md

@@ -35,6 +35,12 @@ chunks.
 Provider snapshots shown in the client are decorated with ingestion coverage
 and status.
 
+NeoRecall is intentionally queried on demand instead of being copied into
+NeoAgent memory. Its integration uses a dynamic public OAuth client with PKCE,
+rotating refresh tokens, and read-only scopes. The available tools call
+NeoRecall's local hybrid search and evidence APIs; they cannot access audio or
+trigger NeoRecall's Ask endpoint.
+
 ## Messaging
 
 Messaging providers are long-running channel adapters managed separately from

package/docs/integrations.md

@@ -19,6 +19,8 @@ Connect accounts in **Integrations**.
 | Notion | Pages, databases, blocks, search |
 | Slack | Conversations, history, search, messages |
 | Figma | Files, nodes, rendered images, comments |
+| NeoArchive | Documents, extracted text, metadata, search, upload, archive, reprocessing |
+| NeoRecall | Local hybrid recall search, memories, mini-memories, daily summaries, conversations, transcript evidence |
 | Home Assistant | Entity state and service calls |
 | Trello | Boards, lists, cards, comments, search |
 | Spotify | Playback, history, search, queue |
@@ -48,7 +50,19 @@ Google, Microsoft, GitHub, Notion, Slack, Figma, and Spotify require OAuth appli
    - `GITHUB_CLIENT_ID=your_id_here`
    - `GITHUB_CLIENT_SECRET=your_secret_here`
 
-Home Assistant and Trello can be configured for a user from the application. See Configuration.
+NeoArchive, NeoRecall, Home Assistant, and Trello can be configured for a user from the application. See Configuration.
+
+### NeoRecall
+
+Open **Integrations**, choose **NeoRecall**, and enter the base URL of the
+NeoRecall server. NeoAgent validates the server and opens NeoRecall's own
+authorization page. Sign in there and approve the three read-only scopes.
+
+NeoRecall contributes seven on-demand tools for local hybrid search and for
+reading memories, mini-memories, daily summaries, conversations, and transcript
+evidence. It does not expose audio, recording, memory mutation, consolidation,
+or NeoRecall Ask. This keeps the products isolated and avoids a second LLM call
+when NeoAgent only needs retrieval context.
 
 ## Messaging channels
 

package/docs/models.md

@@ -44,7 +44,7 @@ Some features use separate providers:
 
 - Embeddings for memory search use a configured supported embedding provider
   and fall back to lexical retrieval when embeddings are unavailable.
-- Recording transcription uses Deepgram when enabled.
+- Voice transcription can use Deepgram when enabled.
 - Image generation and analysis depend on the selected provider and model.
 
 ## Credential handling

package/docs/persistence.md

@@ -13,7 +13,7 @@ The database owns:
 - Tasks, triggers, webhook deliveries, and widgets
 - Integration connections and encrypted credentials
 - Memory, facts, entities, source documents, and retrieval telemetry
-- Recordings, transcripts, health records, and device registrations
+- Health records and device registrations
 - Tool policies and approval history
 
 Larger artifacts and editable agent data use runtime directories under

package/docs/why-neoagent.md

@@ -28,7 +28,7 @@ the current NeoAgent source.
 - You want structured memory and integration ingestion without a separate
   memory service.
 - You need multiple users or specialist agents with separate state.
-- Launcher mode, Health Connect, recordings, or the bundled device interfaces
+- Launcher mode, Health Connect, or the bundled device interfaces
   are part of the use case.
 
 ## Consider OpenClaw when

package/flutter_app/README.md

@@ -14,8 +14,3 @@ A few resources to get you started if this is your first Flutter project:
 For help getting started with Flutter development, view the
 [online documentation](https://docs.flutter.dev/), which offers tutorials,
 samples, guidance on mobile development, and a full API reference.
-
-## Desktop audio capture note
-
-Desktop recording currently depends on `desktop_audio_capture` and is pinned to
-`0.0.2` to avoid unexpected API shifts from automatic minor updates.

package/flutter_app/android/app/src/main/AndroidManifest.xml

@@ -4,10 +4,8 @@
     <uses-permission android:name="android.permission.MANAGE_OWN_CALLS" />
     <uses-permission android:name="android.permission.CAMERA" />
     <uses-permission android:name="android.permission.RECORD_AUDIO" />
-    <uses-permission android:name="android.permission.RECEIVE_BOOT_COMPLETED" />
     <uses-permission android:name="android.permission.FOREGROUND_SERVICE" />
     <uses-permission android:name="android.permission.FOREGROUND_SERVICE_DATA_SYNC" />
-    <uses-permission android:name="android.permission.FOREGROUND_SERVICE_MICROPHONE" />
     <uses-permission android:name="android.permission.health.READ_STEPS" />
     <uses-permission android:name="android.permission.health.READ_HEART_RATE" />
     <uses-permission android:name="android.permission.health.READ_SLEEP" />
@@ -134,18 +132,6 @@
                 android:name="androidx.car.app.minCarApiLevel"
                 android:value="1" />
         </service>
-        <service
-            android:name=".recording.RecordingForegroundService"
-            android:exported="false"
-            android:foregroundServiceType="microphone|connectedDevice" />
-        <receiver
-            android:name=".recording.RecordingBootReceiver"
-            android:enabled="true"
-            android:exported="false">
-            <intent-filter>
-                <action android:name="android.intent.action.BOOT_COMPLETED" />
-            </intent-filter>
-        </receiver>
         <receiver
             android:name=".widgets.AiHomeWidgetProvider"
             android:exported="true">

package/flutter_app/android/app/src/main/kotlin/com/neoagent/flutter_app/MainActivity.kt

@@ -11,15 +11,11 @@ import android.provider.Settings
 import android.provider.OpenableColumns
 import android.view.KeyEvent
 import androidx.activity.result.ActivityResultLauncher
-import androidx.activity.result.contract.ActivityResultContracts
-import androidx.core.content.ContextCompat
 import androidx.core.content.FileProvider
 import androidx.health.connect.client.PermissionController
 import androidx.lifecycle.lifecycleScope
 import com.neoagent.flutter_app.health.HealthConnectGateway
 import com.neoagent.flutter_app.health.HealthSyncScheduler
-import com.neoagent.flutter_app.recording.RecordingForegroundService
-import com.neoagent.flutter_app.recording.RecordingStateStore
 import com.neoagent.flutter_app.widgets.AiHomeWidgetProvider
 import com.neoagent.flutter_app.widgets.AiWidgetStore
 import com.neoagent.flutter_app.widgets.VoiceLaunchWidgetProvider
@@ -46,12 +42,8 @@ class MainActivity : FlutterFragmentActivity() {
     private lateinit var healthGateway: HealthConnectGateway
     private lateinit var healthSyncScheduler: HealthSyncScheduler
     private lateinit var widgetSyncScheduler: WidgetSyncScheduler
-    private lateinit var recordingStateStore: RecordingStateStore
     private lateinit var permissionLauncher: ActivityResultLauncher<Set<String>>
-    private lateinit var microphonePermissionLauncher: ActivityResultLauncher<String>
     private var pendingPermissionResult: MethodChannel.Result? = null
-    private var pendingRecordingResult: MethodChannel.Result? = null
-    private var pendingRecordingArgs: Map<*, *>? = null
     private var launcherButtonSink: EventChannel.EventSink? = null
     private var widgetEventSink: EventChannel.EventSink? = null
     private var appLaunchEventSink: EventChannel.EventSink? = null
@@ -76,7 +68,6 @@ class MainActivity : FlutterFragmentActivity() {
                 .build()
             telecomManager.registerPhoneAccount(phoneAccount)
         }
-        recordingStateStore = RecordingStateStore(this)
         permissionLauncher = registerForActivityResult(
             PermissionController.createRequestPermissionResultContract(),
         ) {
@@ -86,32 +77,6 @@ class MainActivity : FlutterFragmentActivity() {
                 pending?.success(buildStatusMap())
             }
         }
-        microphonePermissionLauncher = registerForActivityResult(
-            ActivityResultContracts.RequestPermission(),
-        ) { granted ->
-            val pending = pendingRecordingResult
-            val args = pendingRecordingArgs
-            pendingRecordingResult = null
-            pendingRecordingArgs = null
-            if (!granted) {
-                pending?.error(
-                    "recording_permission_denied",
-                    "Microphone permission is required for background recording.",
-                    null,
-                )
-                return@registerForActivityResult
-            }
-            try {
-                startRecordingService(args)
-                pending?.success(recordingStateStore.statusMap())
-            } catch (err: Exception) {
-                pending?.error(
-                    "recording_start_failed",
-                    err.message ?: err.javaClass.simpleName,
-                    null,
-                )
-            }
-        }
         MethodChannel(
             flutterEngine.dartExecutor.binaryMessenger,
             "neoagent/telecom",
@@ -258,56 +223,6 @@ class MainActivity : FlutterFragmentActivity() {
             }
         }
 
-        MethodChannel(
-            flutterEngine.dartExecutor.binaryMessenger,
-            "neoagent/recordings",
-        ).setMethodCallHandler { call, result ->
-            when (call.method) {
-                "status" -> result.success(recordingStateStore.statusMap())
-
-                "startBackgroundRecording" -> {
-                    try {
-                        val args = call.arguments as? Map<*, *>
-                        if (ContextCompat.checkSelfPermission(
-                                this,
-                                android.Manifest.permission.RECORD_AUDIO,
-                            ) == android.content.pm.PackageManager.PERMISSION_GRANTED
-                        ) {
-                            startRecordingService(args)
-                            result.success(recordingStateStore.statusMap())
-                        } else {
-                            pendingRecordingResult = result
-                            pendingRecordingArgs = args
-                            microphonePermissionLauncher.launch(android.Manifest.permission.RECORD_AUDIO)
-                        }
-                    } catch (err: Exception) {
-                        result.error(
-                            "recording_start_failed",
-                            err.message ?: err.javaClass.simpleName,
-                            null,
-                        )
-                    }
-                }
-
-                "pauseBackgroundRecording" -> {
-                    startService(RecordingForegroundService.buildPauseIntent(this))
-                    result.success(recordingStateStore.statusMap())
-                }
-
-                "resumeBackgroundRecording" -> {
-                    startService(RecordingForegroundService.buildResumeIntent(this))
-                    result.success(recordingStateStore.statusMap())
-                }
-
-                "stopBackgroundRecording" -> {
-                    startService(RecordingForegroundService.buildStopIntent(this))
-                    result.success(recordingStateStore.statusMap())
-                }
-
-                else -> result.notImplemented()
-            }
-        }
-
         MethodChannel(
             flutterEngine.dartExecutor.binaryMessenger,
             "neoagent/app_update",
@@ -613,19 +528,6 @@ class MainActivity : FlutterFragmentActivity() {
         return super.onKeyUp(keyCode, event)
     }
 
-    private fun startRecordingService(args: Map<*, *>?) {
-        val backendUrl = args?.get("backendUrl")?.toString().orEmpty()
-        val sessionCookie = args?.get("sessionCookie")?.toString().orEmpty()
-        val sessionId = args?.get("sessionId")?.toString().orEmpty()
-        val intent = RecordingForegroundService.buildStartIntent(
-            this,
-            backendUrl = backendUrl,
-            sessionCookie = sessionCookie,
-            sessionId = sessionId,
-        )
-        ContextCompat.startForegroundService(this, intent)
-    }
-
     private fun currentAppMode(): String {
         return try {
             val appInfo = packageManager.getApplicationInfo(

package/flutter_app/lib/main.dart

@@ -17,7 +17,6 @@ import 'package:flutter_svg/flutter_svg.dart';
 import 'package:file_picker/file_picker.dart';
 import 'package:flutter_secure_storage/flutter_secure_storage.dart';
 import 'package:google_fonts/google_fonts.dart';
-import 'package:image/image.dart' as img;
 import 'package:mobile_scanner/mobile_scanner.dart';
 import 'package:qr_flutter/qr_flutter.dart';
 import 'package:shared_preferences/shared_preferences.dart';
@@ -39,8 +38,6 @@ import 'src/health_bridge.dart';
 import 'src/live_voice_capture.dart';
 import 'src/messaging_access_summary.dart';
 import 'src/oauth_launcher.dart';
-import 'src/recording_bridge.dart';
-import 'src/recording_payloads.dart';
 import 'src/security/password_strength.dart';
 import 'src/stream_renderer.dart';
 import 'src/theme/palette.dart';
@@ -66,7 +63,6 @@ part 'main_timeline.dart';
 part 'main_runtime.dart';
 part 'main_controller.dart';
 part 'main_devices.dart';
-part 'main_recordings.dart';
 part 'main_chat.dart';
 part 'main_account_settings.dart';
 part 'main_settings.dart';
@@ -104,8 +100,6 @@ const String _neoMailLogoAsset = 'assets/branding/neomail_logo.svg';
 const String _sessionCookiePrefsKey = 'auth.sessionCookie';
 const String _sessionCookieBackendPrefsKey = 'auth.sessionCookieBackend';
 const String _sessionCookieSecureStorageKey = 'auth.sessionCookie.secure';
-const int _voiceAssistantScreenshotMaxDimension = 1600;
-const int _voiceAssistantScreenshotMaxBytes = 900 * 1024;
 
 String get _desktopTrayIconAsset =>
     defaultTargetPlatform == TargetPlatform.macOS

package/flutter_app/lib/main_app_shell.dart

@@ -2521,8 +2521,6 @@ class _SectionBody extends StatelessWidget {
         return VoiceAssistantPanel(controller: controller);
       case AppSection.devices:
         return DevicesPanel(key: devicesPanelKey, controller: controller);
-      case AppSection.recordings:
-        return RecordingsPanel(controller: controller);
       case AppSection.messaging:
         return MessagingPanel(controller: controller);
       case AppSection.runs: