Android Voice AI SDK (MVVM)
Add a complete voice-driven AI conversation pipeline to any Android app in minutes.
- Source repo
- ahmedeltaher/Agnentic-Voice-AI-SDK
- Stars
- ★ 2.6k
- Last updated
- 5mo ago
- Primary language
- Kotlin
- FA score
- 39/100 · Major gaps
At a glance
- How it runs
- Works with
- Platform-specificClaude API
- Cost
- Free software; you pay for model usage
- Setup effort
- Medium · a few setup steps
- You'll need
- Typical use
- Mobile developers wanting to add a voice assistant to their Android app quickly without building STT/AI/TTS pipelines from scratch.
- Not a fit if
- Teams needing cross-platform iOS, web, or desktop voice support
- Projects unwilling to pay for and supply an Anthropic API key
- Website
- ahmedeltaher.github.io ↗ set on GitHub
- Maintenance
- Commits in the last 6 monthslast commit 2026-06-01
- Source review
- 39/100 · Major gaps 1 safety controls not found
What does this agent do, and when should you use it?
The Android Voice AI SDK is a reusable Android library built on MVVM that provides a full voice-driven AI conversation pipeline. The core pipeline captures microphone audio, detects voice activity, transcribes speech to text (STT), sends the transcript to Anthropic Claude for an intelligent response, and speaks the reply back through text-to-speech (TTS). The SDK ships ready-to-use Jetpack Compose UI components (VoiceButton, ConversationView), swappable STT/TTS engine adapters (Android built-in, OpenAI Whisper, ElevenLabs), on-device emotion detection, and security utilities including PII redaction and encrypted key storage. It uses Hilt for dependency injection and supports building voice sessions, configuring models, and system prompts. The library requires Android 7.0 (API 24) and an Anthropic API key.
The SDK orchestrates a voice conversation pipeline: it captures audio via AudioRecord, detects speech with built-in VAD, and transcribes using a SpeechToTextEngine (default Android SpeechRecognizer, optional Whisper). The text is sent to an AIEngine (ClaudeAIEngine wrapping the official Anthropic Java SDK) which generates a response while maintaining conversation history. Finally, a TextToSpeechEngine (default Android TTS, optional ElevenLabs) synthesizes and plays the reply. The VoiceAISession orchestrates the flow with start() and stop() methods. Integration steps include adding the dependency, declaring permissions, setting up Hilt (@HiltAndroidApp, @AndroidEntryPoint), providing the API key via local.properties, and building with VoiceAISDK.Builder. The UI uses VoiceSessionPermissionGate for runtime permission and provides composables like VoiceButton and ConversationView.
- Mobile developers wanting to add a voice assistant to their Android app quickly without building STT/AI/TTS pipelines from scratch.
- Apps needing flexible STT/TTS engines, switching between Android built-in and OpenAI Whisper or ElevenLabs for different accuracy and voice quality.
- Applications handling sensitive data that require PII redaction to strip phone numbers, emails, and credit cards before sending to AI.
- Apps that want emotion-aware responses, enabling on-device emotion detection to let Claude adapt its tone per detected emotion.
- Security-conscious apps that want encrypted API key storage via Android Keystore and certificate pinning for network requests.
Key setup facts
Pulled from the install and usage notes below. The project README has the current values.
- Environment variables
ANTHROPIC_API_KEY- Install notes cover
- Android
How do you install or deploy this agent?
Add the dependency in app/build.gradle.kts: implementation("com.sdk:voice-ai-sdk:1.0.0"). Declare permissions in AndroidManifest.xml: RECORD_AUDIO, INTERNET, ACCESS_NETWORK_STATE. Set ANTHROPIC_API_KEY in local.properties and expose it via buildConfigField. Use Android Studio Meerkat or newer, and set up Hilt for dependency injection (optional but recommended).
How do you use this agent?
Annotate your Application class with @HiltAndroidApp and Activity with @AndroidEntryPoint. Create an AppModule to provide VoiceAIConfig and VoiceAISDK, building the SDK with the API key. In Compose, wrap VoiceButton and ConversationView with VoiceSessionPermissionGate. Obtain a VoiceAISession and call start() to begin conversations. Configure options in the config { } block, such as systemPrompt, inputMode, silenceTimeoutMs, and piiRedaction.
What are this agent's strengths and limitations?
- Complete voice pipeline with pluggable STT, TTS, and AI engines, offering flexibility.
- Ready-to-use Jetpack Compose UI components simplify integration into modern Android apps.
- MVVM architecture ensures clean separation of concerns and testability.
- Security features like PII redaction, encrypted key storage, and certificate pinning protect sensitive data.
- On-device emotion detection works without extra API keys, enhancing interaction.
- Clear documentation with code examples for quick start.
- Android-only; not suitable for cross-platform projects.
- Core AI relies on Anthropic Claude API, requiring a paid API key and incurring usage costs.
- Documentation lacks specifics like exact Gradle plugin versions, potentially causing integration friction.
- Additional features (Whisper, ElevenLabs, Hume AI) require separate API keys and network access.
- On-device emotion detection has limited accuracy compared to cloud-based solutions.
- Unclear license status in repository metadata (though README states MIT).
How does this agent compare with similar options?
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Android Voice AI SDK (MVVM) This agent | 39 · Major gaps | Library / SDKFree + model costs | ★ 2.6k | 5mo ago | Kotlin | Claude API |
| Voice AI Learning Path | 68 · Some gaps | Library / SDKFree | ★ 341 | 19d ago | Astro | — |
| Patter Voice SDK | 65 · Some gaps | Library / SDKFree + model costs | ★ 1.1k | 2mo ago | Python | OpenAI API · Claude API |
| Stream Vision Agents | 57 · Major gaps | Library / SDKFree + model costs | ★ 8.2k | today | Go | OpenAI API · Claude API |
How does FollowAgents rate this agent?
Why each dimension lost points
The README describes microphone capture, network requests, and optional emotion analysis, and documents a runtime microphone permission gate, PII redaction, encrypted storage, and certificate pinning options, making data flows reasonably visible (2). However, permission scope and external calls are not bounded by scenario; emotion consent is left to integrators to configure. The key guidance recommends local.properties/BuildConfig without explaining the exposure risk of embedding a key in a client app, and security controls are mostly asserted. Least privilege, confirmation, sensitive-data handling, and external effects are therefore only partial. No revocation or rollback path, code-level attribution, or clear maintenance owner is shown.
The README organizes the product around audio, STT, AI, TTS, session, and UI, and lists dependency versions and engines, which sketches an ordinary integration path (dependency availability 1). It claims a broad set of capabilities while detailing Claude and only some default engines; package coordinates and version claims are not corroborated by other files, limiting self-consistency (1). The supplied material gives no error types, failure messages, or recovery behavior, so failure_messages is 0.
The README targets developers adding voice conversations to Android apps, provides Compose/Hilt and direct-construction paths, and identifies built-in and cloud engines, giving useful audience and environment guidance (2 each). Capability boundaries remain unclear: the feature table lists intent/entity extraction, workflows, and analytics without implementation detail or limits. Trigger behavior is described as default HANDS_FREE and optional push-to-talk, with little about background, lifecycle, or trigger boundaries, so those criteria score 1.
The README includes a layered architecture, configuration table, quick start, and copyable examples, supporting good information architecture and installation guidance (2 each). Only the README is supplied, so naming and API stability cannot be cross-checked; there is no FAQ and limitations are only briefly indicated through network and key requirements. MIT is stated in the README, but license metadata is unknown and no repository license file is provided. A 1.0 label appears without a change history. No maintenance owner or update path is identified (0).
If implementation matches the documentation, a single session pipeline, swappable engines, and Compose components can reduce integration work, and the README supplies usable integration examples, supporting output_usability and marginal_value at 2. Cost-benefit is only partly covered: multiple cloud services require separate keys, while latency, pricing, quotas, and operating costs are not discussed, so cost_benefit is 1.
The README's architecture, configuration, and setup steps make some claims traceable (1), but no source code, tests, build configuration, or second source is supplied for comparison, so cross-source corroboration is 0. The document distinguishes some configuration fields from their intended effects, but many feature and security claims lack a verification path; fact and inference are only partly separated (1).
- Not found in source: rollback or recovery pathBack up first, or work on a git branch or snapshot, so its changes can be undone.
- Do not treat a BuildConfig key in a mobile app as a secret credential; assess client extraction risk and whether a server-side key proxy is needed before deployment.
- Emotion analysis and voice data can be sensitive; confirm which engines are enabled, what data is transmitted, the consent flow, and retention periods.
- The listed capabilities, dependency coordinates, and MIT license are not corroborated by source code, build files, or an independent license file in this packet.
FAQ
Is an Anthropic API key required?
What Android versions are supported?
Can I use the SDK without Hilt?
How is audio privacy handled?
Can I switch the STT/TTS engines?
Is Android Voice AI SDK (MVVM) free?
Is Android Voice AI SDK (MVVM) open source?
Is Android Voice AI SDK (MVVM) still maintained?
Who makes Android Voice AI SDK (MVVM)?
What is Android Voice AI SDK (MVVM)'s official website?
Thanks — we use this to improve the page.