Why I Built Enclavi: Wanting AI Productivity in Regulated Environments Without Leaking Client Secrets

Published on: 2026-10-06
5 min read
Author: Sidi Mohamed Idrissi Yaghir

For years, I have worked in highly regulated environments. My clients come from industries where data confidentiality is not a vague talking point on a corporate webpage—it is enforced through comprehensive non-disclosure agreements, compliance audits, and legal oversight.

In this day-to-day work, meetings and multi-hour workshops are standard: strategy roadmaps, architectural design reviews, process analyses, and executive decision-making.

Anyone who facilitates these workshops knows the exhaustion that follows: you wrap up three or four hours of intensive dialogue. Whiteboards are covered in diagrams and post-it notes, and your desk is piled with handwritten bullet points that must somehow be synthesized into an actionable, professional summary.


The Dilemma: Massive Leverage, Unacceptable Risk

Roughly two years ago, the generative AI boom arrived. Suddenly, tools appeared that could capture audio, generate transcripts, and organize action items automatically.

I tested these tools in personal settings and was impressed: the reduction in routine cognitive labor was enormous. Instead of spending two hours in the evening piecing together quotes, you have a solid first draft ready in minutes.

Yet in client engagements, I was strictly barred from using any of them.

Every existing service on the market followed the identical operational playbook:

  1. You hit record.
  2. The audio file is streamed to servers hosted on AWS, Google, or Azure.
  3. Transcription pipelines and third-party LLM inference APIs process the payload remotely.

Even when vendors promise data is "encrypted in transit" and "not used for model training," uploading recordings during a confidential workshop represented a clear violation of my contractual duties and professional confidentiality.

I faced an uncomfortable trade-off: either forego modern productivity tools entirely, or gamble with the trust my clients placed in me.


"I Will Just Build It Myself"

I refused to accept that compromise. Having followed modern processor design, I knew how capable Apple's M-series and A-series chips had become. If an iPhone can perform trillions of operations per second on its Neural Engine, why should speech recognition models like Whisper not run directly on the device?

So I began building Enclavi.

My intention was never to create just another voice memo app. The goal was to build a rigorous professional tool that pairs uncompromising privacy with real practical utility for knowledge workers.

Sovereign Architecture

The Sovereign Enclavi Workflow

From raw spoken voice to structured document entirely on device

1. Conversation / Workshop

High-fidelity recording via iPhone studio mics. Airplane mode supported.

2. Apple Neural Engine (ANE)

100% on-device WhisperKit transcription. Zero bytes sent to cloud servers.

3. Document-First Canvas

Instant structuring into executive notes, action items, Markdown export & local vector search.


How Enclavi Differs: Core Design Principles

Derived directly from the realities of my client engagements, Enclavi is built on four non-negotiable principles:

  1. No Accounts, No Tracking, No Servers: Enclavi requires no login or registration. There is no telemetry, no tracking beacon, and no remote server backend. The application works completely in Airplane Mode.
  2. On-Device Whisper Transcription: Audio is converted to text using optimized CoreML weights executing on your iPhone's Neural Engine. The audio never leaves your physical hardware.
  3. Document-First Canvas: An audio recording is merely raw material, not the deliverable. Enclavi translates spoken dialogue into an editable canvas with executive takeaways, decisions, and transcripts ready for immediate use.
  4. Absolute Data Sovereignty: Everything is stored in your local on-device database. You decide whether to export your notes as Markdown or integrate them with your existing local note archives.

Real Trust Requires No Terms of Service

Today, when I lead a client workshop, I can place my phone on the table, open Enclavi, and focus entirely on the discussion.

I do not have to ask clients for permission to connect a cloud bot, nor do I leave the session wondering where the audio stream will end up.

True trust is not earned through promises written in a privacy policy. It is built by choosing software architectures that make data leakage physically impossible.