Your Lenn, at home

What you need

• A server with Docker and Docker Compose: a small rented server, a computer at home, a recent NAS.
• Memory: the compose file gives 1 GB to the web app and 4 GB to the worker that transcribes and understands your moments. Plan for about 6 GB. The optional audio service needs more; an NVIDIA GPU speeds it up.
• An HTTPS address, for example lenn.yourdomain.com, behind a reverse proxy. A Caddy example is included.
• Your keys at the AI providers, or nothing at all to start: demo mode works without any key.

Install in five steps

1. 1

   Get the code

   Clone the repository on your server.

   $ git clone https://github.com/Synapsr/lenn.git && cd lenn
   

2. 2

   Fill in three values

   Copy the example file, then set LENN_PUBLIC_URL (your server’s public address), LENN_AUTH_SECRET and POSTGRES_PASSWORD. Never change the secret afterwards: your saved keys and devices depend on it.

   $ cp deploy/.env.example deploy/.env
   $ openssl rand -base64 48    # LENN_AUTH_SECRET
   $ openssl rand -hex 24       # POSTGRES_PASSWORD
   

3. 3

   Start Lenn

   One command starts PostgreSQL, applies the migrations, then runs the web app and the worker. The app listens on 127.0.0.1:3210, never directly on the internet.

   $ docker compose -f deploy/compose.yaml up -d --build
   

4. 4

   Put it behind HTTPS

   Place a reverse proxy in front of Lenn. With Caddy, the certificate is automatic. Add LENN_TRUST_PROXY=true to deploy/.env. See the full example.

   lenn.example.org {
     reverse_proxy 127.0.0.1:3210 {
       flush_interval -1
     }
   }
   

5. 5

   Open the assistant

   Open your address in a browser. The first visit creates the admin account, then the assistant helps you paste your keys and test them with one button. They are encrypted in the database; environment variables always take precedence.

Your keys, and what they do

Lenn only shows what is really configured on your server: without an OpenAI key, for example, there is no voice conversation button. You pay each provider directly, at their prices.

With no key at all

Add LENN_AI=mock to deploy/.env: fake transcripts and answers, perfect to try it out, give a demo or contribute to the code.

Other providers, local ones included

• Transcription: Deepgram, or any OpenAI-compatible Whisper server, including faster-whisper at home.
• Companion: any OpenAI-compatible API, such as Mistral, or a local model with Ollama.

The audio service (optional)

A Python service that runs next to Lenn: noise reduction, speech detection, speaker separation with pyannote, and voiceprints to recognise who is speaking. It runs on CPU, faster with a GPU. Without it, Lenn still transcribes your moments.

$ docker compose -f deploy/compose.yaml -f deploy/compose.audio.yaml up -d --build

Then set LENN_AUDIO_URL=http://audio:8000 and LENN_AUDIO_TOKEN in deploy/.env. The Hugging Face token requires accepting the pyannote model conditions once.

Licences

The code is on GitHub: server licence, Apple apps licence.

Contribute

• A problem, an idea: open an issue.
• The design system describes the companion, colours, voice and tone: one product everywhere.
• Don’t feel like installing anything? Lenn Cloud does all of this for you.