Starlog
Security and architecture

How Starlog keeps your conversations private.

From the phone on the table to the summary in your pocket: where your words go, who handles them, and what keeps each company apart. Only what runs today is drawn as there.

One meeting, step by step

Where your words go.

  1. Recorded on the phone

    The audio is saved on the phone first, so a bad signal loses nothing, and sent in small encrypted parts.

  2. Received in Germany

    Starlog's server keeps it with the meeting and checks who may see it before anything else happens.

  3. Turned into words

    Speech-to-text services write the words and tell the voices apart. Training is off; every phone that heard the meeting is used.

  4. Summarised

    Claude writes the summary, the decisions and the tasks. Anthropic does not train on what we send.

  5. Filed where it belongs

    The meeting lands in the right company's space, the tasks with their owners, and the push goes to the people who were there.

Secure meetings

A meeting only the people in it can open.

On Plus and up, turn on Secure before you record. Others join by scanning a code on your phone, and every phone records secure.

  • Once the meeting is ready, its transcript and summary are encrypted with keys made from the passkeys of the people in it.
  • Nobody else can open it, not the company's other members, not us.
  • The audio of every phone is deleted as soon as it is transcribed, and never goes into a backup.
  • The host can share it with one more person; everyone in the meeting sees who was added, and by whom.
Honest about the limit: while a secure meeting is being transcribed and summarised, our server and the services above see its words. They are sealed the moment it is ready.
The stack

What Starlog is made of.

The appA web app you install from the browser on any phone or computer, written in TypeScript and React. Apps for iPhone and Android are on the way.
The serverTypeScript on Node.js, PostgreSQL, and one container per company, at Hetzner in Falkenstein, Germany. Audio older than a week moves to storage in the same place; backups go to Helsinki, Finland.
Speech to textpyannoteAI (France) and ElevenLabs for the words and who spoke when, with training switched off at each. Kami, our own engine on our own hardware, is being built to replace them.
AIClaude by Anthropic writes summaries and answers. Each company has its own AI agent, built on the Claude Agent SDK, that works only inside its sealed space.
Sign-inPasskeys on your phone: no passwords to steal or reuse. A lost phone comes back through your confirmed email.
EncryptionTLS on every connection. Secure meetings sealed with keys from their participants' passkeys (AES-256-GCM).
Your AIA personal, read-only key for your own AI through MCP, which you can replace or remove at any time.

Every company that handles your data for us, and where: sub-processors. What we keep and your rights: privacy policy.

Being built

Before Starlog opens to everyone.

Our own transcription Coming

Kami, our own speech-to-text on our own hardware, so audio no longer leaves our servers. The meeting will say which engine wrote it.

A key per company Coming

Every company's meetings encrypted at rest with a key of its own.

Servers in Switzerland Coming

Starlog is made by a Swiss company; its servers follow.

Your phone is already ready.

Starlog opens in small groups. Join the waiting list, and start with 30 days free when it is your turn.

Join the waiting list