No Backend Personal Project Direct API Calls Android Dictation Local AI PoC

Your API Key.
Your Intelligence.

EchoScribe started as a private tool for my own daily summaries and dictation. I now share it with others who want modern AI features without giving an app provider their data, account, or hidden backend access.

Play Store Web-App
OpenAI Gemini Claude Grok

AI Image Generation

Create visuals directly through GPT Image 2, Gemini 3 Image, or xAI Grok Imagine. EchoScribe returns the generated image in-app so you can inspect, share, or save it.

Transcription

Record speech and convert it to text with OpenAI, Gemini, xAI Grok, or a local Whisper-compatible endpoint. OpenAI Realtime can stream live text while recording through gpt-realtime-whisper.

Floating Dictation

On Android, a movable system-wide dictation button appears only in safe editable fields. Dictate messenger replies, clean up raw speech into polished text, or draft formal letters before inserting only after approval.

Summarization

Turn transcripts, pasted text, shared messages, or webpages into concise summaries using OpenAI, Gemini, Claude, Grok, or your own Ollama-compatible Local AI endpoint.

Intelligent Re-Translation

Switch target language on the fly. EchoScribe can re-process the original source after recording, or stream live OpenAI Realtime Translation through gpt-realtime-translate.

Neural TTS

Listen to generated summaries through OpenAI, Gemini, or xAI TTS. Playback is cached locally for quick replay.

100% Private

No EchoScribe servers, no account system, no tracking. Your phone, desktop companion, or browser extension talks directly to the provider or local endpoint configured in settings.

Android app

Capture, summarize, configure, and reuse from one compact interface

The mobile app keeps the everyday workflow close to the thumb: record or paste content, switch between transcription and summary, choose the provider, and return to local history when you need a result again.

  • Start with a clean transcription panel and a single record action.
  • Switch to summaries without leaving the current source material.
  • Configure OpenAI, Gemini, Claude, Grok, Local AI, prompts, and Floating Dictation in settings.
EchoScribe Android transcription view
Transcription
EchoScribe Android summary mode
Summary
EchoScribe Android API settings and provider configuration
API Config
EchoScribe Android local history view
History

Built for everyday capture

From source to useful output without a backend

EchoScribe is designed around quick capture and reuse. You can record an idea, share a page from the browser, paste a chat message, dictate a reply, turn rough speech into clean writing, draft a formal letter, translate the result, and copy or listen to the final text without sending anything through an EchoScribe-owned server.

Audio notes and meetings

Record directly in the app, transcribe through your selected provider, then generate a structured summary. With OpenAI Realtime enabled, EchoScribe shows live transcription while you speak; without it, the app keeps the reliable finished-recording transcription workflow.

Android Floating Dictation

Use EchoScribe outside the app as a movable dictation bubble for chat replies, clean rewrites, and formal wording help. Password, PIN, payment, phone, and banking fields stay blocked; EchoScribe never sends messages or taps buttons automatically.

URLs and webpages

Share a URL to EchoScribe or paste it from the clipboard. The app extracts readable page content locally before sending text to the AI provider, which keeps behavior predictable for providers that do not fetch URLs themselves.

Realtime Translation

Choose a target language and EchoScribe streams spoken audio to OpenAI's dedicated Realtime Translation model. The result appears as translated text during recording, and the translation mode is constrained to translate questions instead of answering them.

Share intent and clipboard flow

Send text, links, or compatible audio from other Android apps straight into EchoScribe. Recent fixes make re-sharing a cleared or history-deleted URL start a fresh extraction instead of reopening stale empty content.

History, fullscreen, and reuse

Local history keeps transcripts and summaries on your device for later reuse. Long results can be opened in fullscreen reading mode, copied, shared, translated again, or played back as speech.

Prompt and model control

Advanced users can adjust summary prompts and switch providers or Pro mode in settings. EchoScribe keeps model defaults current while still letting you decide which provider receives each request.

Debug visibility

Optional debug logging helps inspect provider requests and troubleshoot API-key or response issues. It is intended for power users and remains local to the app session.

Latest release: EchoScribe 1.0.49 adds Claude Opus 5 and hardware-aware Local AI model recommendations to the Linux and Windows installers. The published packages were rebuilt with privacy-clean deterministic paths and without public Windows debug symbols.

Desktop companions

Push-to-talk dictation and browser summaries on Windows and Linux/GNOME

EchoScribe is still Android-first, but the same BYOK approach now extends to optional desktop companions. Desktop requests go directly from your computer to the provider configured by you. There is no EchoScribe relay server in the middle.

Windows

Install the tray app with install.cmd. The setup wizard can choose the install folder, create Start Menu and per-user autostart entries, register the local Native Messaging host, and optionally configure native Windows whisper.cpp and Ollama.

Linux/GNOME

Run ./install.sh from the Linux package. The wizard installs dependencies, prepares the Python environment, configures provider keys, installs the GNOME Shell extension, and registers browser Native Messaging hosts. CUDA/Whisper and Ollama remain optional.

Browser summaries

Chrome, Edge, Brave, Chromium, and Firefox are supported through local browser extensions. The summary popup can use automatic language matching or a fixed target language before sending the request to the Native Messaging host.

Downloads and setup guide

Desktop packages are published as separate Windows and Linux/GNOME assets on GitHub releases. The repository README contains the current setup instructions for Android, Windows, Linux/GNOME, autostart, browser extension loading, uninstall, and release checksums.

EchoScribe browser extension popup with provider and language controls
Extension popup with provider, model, language, summarize, and copy controls.
EchoScribe browser extension summarizing a webpage
Webpage summary returned inside the browser, through the local Native Messaging host.
Chrome extensions page loading the EchoScribe extension manually
Manual extension loading keeps browser installation explicit and user-controlled.

Browser summaries

Summarize the page you are already reading

The Chromium and Firefox extensions keep the browser workflow local. They collect the current page content, ask the desktop Native Messaging host to process it with your configured provider, and return the summary inside the popup.

  • Use automatic language matching or choose a fixed target language.
  • Copy the generated result directly from the extension popup.
  • Load the browser extension manually while the installer registers the native host.

Linux/GNOME

GNOME integration for push-to-talk and local settings

On GNOME, EchoScribe lives close to the desktop shell. The extension exposes quick access to settings while the companion app handles provider configuration, dictation behavior, and the local bridge used by the browser extensions.

  • Push-to-talk dictation is built for fast desktop text input.
  • Settings stay local to the machine and can be adjusted without a cloud account.
  • Windows companion visuals will be added here once the Windows app screenshots are ready.
GNOME extensions view showing EchoScribe settings entry
GNOME extension entry point for opening EchoScribe settings.
EchoScribe GNOME settings dialog
GNOME settings surface for desktop companion behavior.
Desktop setup: download the current Windows or Linux/GNOME companion package from GitHub Releases, then follow the README install and browser-extension loading steps. Download v1.0.49+57

Optional Local AI PoC

Run summaries and dictation through your own LAN backend

Local AI is for users who want to experiment with fully self-hosted text summaries, translations, and speech-to-text. EchoScribe expects an Ollama-compatible /api/chat endpoint and an OpenAI-compatible Whisper transcription endpoint. Keep both on a trusted LAN or VPN; EchoScribe does not add authentication to Local AI requests.

1

Install

Install Ollama plus Python, FastAPI, Uvicorn, faster-whisper, ffmpeg, and zstd on the Local AI host.

2

Pull models

The Linux and Windows installers compare CPU, RAM, GPU, and VRAM with built-in model profiles. They can optionally request additional CanIRun.ai suggestions after showing the hardware values that would be sent; an offline-only mode remains available.

3

Start services

Run Ollama on 0.0.0.0:11434 and the Whisper server on 0.0.0.0:8000.

4

Configure EchoScribe

Select Local AI for STT or summaries and use the host LAN IP from Android, Windows, or Linux/GNOME.

Linux host

sudo apt update
sudo apt install curl ffmpeg python3 python3-venv python3-pip zstd
ollama pull qwen3.5:9b
export OLLAMA_HOST=0.0.0.0:11434
ollama serve

Whisper server

python3 -m venv .venv
. .venv/bin/activate
python -m pip install --upgrade pip
python -m pip install fastapi "uvicorn[standard]" python-multipart faster-whisper
export WHISPER_MODEL_SIZE=large-v3-turbo
export WHISPER_DEVICE=cpu
export WHISPER_COMPUTE_TYPE=int8
uvicorn whisper_server.app:app --host 0.0.0.0 --port 8000

Windows native

The Windows companion can configure local summaries with Ollama for Windows and local speech-to-text with whisper.cpp. CUDA is used only when compatible NVIDIA support is already present; EchoScribe does not install NVIDIA drivers or CUDA toolkits.

Advanced WSL option

WSL is no longer the default Windows path. Advanced users can still point EchoScribe at an existing WSL or LAN Whisper endpoint when they already maintain that backend themselves.

EchoScribe settings: Local AI LLM URL http://host-ip:11434/api/chat, a selected model such as qwen3.5:9b, Whisper URL http://host-ip:8000/v1/audio/transcriptions for server backends, or native Windows whisper.cpp paths configured by setup.

Why EchoScribe exists

Built for myself first, shared because privacy should be practical

There are many apps that can summarize, transcribe, translate, or generate content. The problem for me was the usual model behind them: accounts, subscriptions, opaque backends, and no real bring-your-own-key principle. In that setup, users often have to trust not only the AI provider, but also the app operator in the middle.

I wanted the opposite for my own use: no EchoScribe backend, no subscription gate, no extra provider account controlled by the app, and no hidden place where recordings, URLs, prompts, transcripts, or summaries could be collected. So I built EchoScribe as a BYOK app and now make it available to others who want the same direct-control approach.

Your provider account

You use your own OpenAI, Gemini, Claude, or xAI key, or route supported text and STT work to your own Local AI endpoints.

No middleman backend

EchoScribe does not proxy AI requests through an app-provider server that could inspect or retain your content.

Personal tool, public app

It began as a private workflow for myself and is now offered openly for users who prefer transparent BYOK software.

Elite AI Performance

Switch between Standard and Pro models with a single toggle in the settings. Standard favors speed and cost control; Pro uses the strongest available model choices for demanding summaries, translation, and image generation. Local AI remains separately configurable by endpoint and model name.

Standard Models

  • GPT-5.6 Terra
  • Gemini 3.6 Flash
  • Claude Sonnet 5
  • Grok 4.3
  • Local AI (configurable)
  • gpt-realtime-whisper
  • gpt-realtime-translate

Pro Models

  • GPT-5.6 Sol
  • Gemini 3.1 Pro Preview
  • Claude Opus 5
  • Grok 4.5
Switch in Settings

BYOK Architecture

EchoScribe is a high-performance bridge, not a middleman. There is no EchoScribe account system, API relay, cloud workspace, sync service, analytics backend, or hidden server-side processing layer. Your encrypted API key stays on your device and each AI request is sent directly from your phone or browser to the provider you select.

1

Bring Your Key

Paste your OpenAI, Google, Claude, or xAI key, or configure Local AI endpoints

2

Hardened Storage

AES encryption in Android Keystore

3

Pure Privacy

No backend logs, no tracking, no ads

Privacy policy summary

What happens to your data

EchoScribe is built around local control. The app can process recordings, shared audio, pasted text, URLs, prompts, transcripts, summaries, translations, generated images, TTS audio, local history, settings, API keys, and optional debug logs, but EchoScribe does not collect those contents on an EchoScribe server. They remain on your device unless you explicitly start a provider request, share an output, export/save a file, or use the public website.

No EchoScribe backend for AI data

There are no EchoScribe servers that receive, store, review, train on, or forward your recordings, prompts, API keys, transcripts, summaries, translations, generated images, or TTS text/audio. The public website only serves the static landing page and web app files; it is not an AI request backend.

Direct provider requests

When you choose OpenAI, Google Gemini, Anthropic Claude, or xAI Grok, EchoScribe sends the required request directly from your end device to that provider's API. When you choose Local AI, supported text and STT requests go to the endpoint you configured. The selected destination receives only the content needed for the feature you started.

Data used only for the feature you request

  • Audio is used for recording, transcription, summaries, and optional dictation workflows.
  • Text, shared messages, extracted webpage text, and custom prompts are used for summaries, translation, and re-processing.
  • Image prompts are used to generate images through the selected image provider.
  • TTS text is used to generate playable audio through the selected speech provider.

Local storage and deletion

API keys are stored locally using platform secure storage such as Android Keystore. History, transcripts, summaries, generated media, cached TTS audio, settings, and debug information are stored locally on your device so you can reuse them. You can delete history, clear app data, remove API keys, or uninstall the app to remove local EchoScribe data. Provider-side retention must be managed in the account or dashboard of the provider you used.

Recipients and third parties

EchoScribe does not sell data, use ad networks, embed tracking SDKs, or share app content with data brokers. AI content is disclosed only to the provider you selected for the action you started. Those providers process the request under their own terms, privacy policies, abuse monitoring, retention rules, and billing settings. EchoScribe cannot delete or change provider-side logs after a direct API call has been made.

Permissions and Android Accessibility

Microphone access is used only when you record. File/media access is used only for saving or sharing results where your platform requires it. The optional Android Floating Dictation feature uses Accessibility only to detect editable fields, show the floating microphone, and insert text after your approval. Password, PIN, payment, banking, credit-card, and phone fields are blocked.

Website and technical access logs

Visiting app.wean.de creates normal technical webserver and Cloudflare delivery/security data, such as IP address, timestamp, requested URL, user agent, TLS/security metadata, and error status. These logs are used for secure delivery, troubleshooting, abuse prevention, and server operation. Current nginx logs rotate daily and are kept for up to 14 rotations. Website logs do not contain your EchoScribe API keys or AI request contents.

Your controls and rights

You decide which provider receives a request, which API key is used, whether Pro mode is enabled, whether history/debug logging is kept locally, and when local data is deleted. Where data protection law such as the GDPR applies, you may request access, correction, deletion, restriction, portability, or objection regarding data controlled by the EchoScribe project. For data processed by OpenAI, Google, Anthropic, or xAI through their API, please contact that provider directly. For EchoScribe project and privacy questions, contact [email protected].