Bygmind

Bygmind is a private AI voice recorder for iOS and Android. Audio capture, transcription, and speaker identification run entirely on your device. No cloud upload, no account required, no telemetry. For the AI reasoning layer — summaries, Q&A over your recordings, PDF report drafting — you point Bygmind at the LLM endpoint of your choice: Claude, GPT, Gemini, or your own local model via Ollama, LM Studio, vLLM, or any OpenAI-compatible inference server. Record meetings, site visits, dictation, walkthroughs, and inspections. Get on-device transcription in 25 languages including German, French, Italian, Spanish, Portuguese, Dutch, Polish, and Russian. Speaker diarization runs locally and re-identifies recurring speakers across recordings. Attach photos and videos. Ask the agent questions about what was said. Draft structured PDF reports from your raw material. Everything stays on the phone unless you actively configure an LLM endpoint. Under the hood, transcription uses Parakeet via the FluidAudio SDK, optimized for the Apple Neural Engine on iOS and equivalent on-device acceleration on Android. Up to 110x real-time on Apple Silicon. The agent loop with tool calling runs on a small local model by default. If you want stronger reasoning, you wire up a cloud endpoint or your own self-hosted LLM via a standard OpenAI-compatible URL and API key in the app settings. What makes Bygmind different from cloud notetakers like Otter, Granola, Fireflies, Plaud, or ChatGPT Record: those products ship your audio to their servers and reserve the right in their privacy policy to train on your conversations. Bygmind keeps audio on the device by structural design, not by feature toggle. Verify it yourself. Turn on airplane mode and the full feature set keeps working. Run it through Little Snitch or a packet capture and you will see the capture and transcription layer makes zero outbound network calls. The only network traffic is whatever LLM endpoint you choose to wire up. If you do not configure one, nothing leaves the device. Real outcomes users can expect: a lawyer or doctor capturing client conversations without violating GDPR or HIPAA constraints because the audio never reaches a third party. An energy consultant or building inspector doing field documentation on-site with no signal, then driving home with a complete transcript and draft report already in their phone. A construction professional dictating walkthrough notes on a noisy site and getting a clean structured PDF report ten minutes later. A founder recording team meetings without their conversations becoming training data for the next foundation model. Key features include on-device speech-to-text in 25 languages, local speaker diarization with recurring speaker re-identification across recordings, an agent loop with tool calling for search and summarize and Q&A and PDF report drafting, BYO LLM endpoint support for Claude, GPT, Gemini, Ollama, LM Studio, vLLM and on-prem stacks, photo and video attachment to recordings, airplane-mode operation with the full feature set offline, no account and no signup required for single users, continuous disk-write recording that survives app kill mid-recording, optional sync to your own Nextcloud instance or S3 bucket, and GDPR compliance by design. Bygmind is built for energy consulting, building inspection, construction documentation, real estate field work, insurance assessment, legal client meetings, medical practitioner notes, architectural site surveys, journalist interviews, academic field research, and accessibility documentation. It is not for phone calls or VoIP. It is for the things you say to your phone: meetings on site, dictation, walkthroughs, inspections. Honest state: iOS is live on the App Store, Android is in open beta on Google Play. Free, no signup required, unlimited recordings and projects. Solo founder, pre-revenue. The app is closed-source for now, but the on-device data flow is verifiable through standard network inspection. Made in Germany.