Self-host.

Run the whole product on your own server. The code is MIT-licensed, and your instance sets its own limits, providers and storage.

Docker Compose.

This brings up Postgres with pgvector, MinIO and the web app. Open the app, and the first-run wizard walks you through the database, storage, sign-in, transcription and LLM. The first user to sign in becomes the admin.

~
$ git clone https://github.com/Heknon/AskCalls.git
$ cd AskCalls/docker
$ docker compose up
# then open http://localhost:3000

Configure.

Every setting is editable at /admin/instance. Any environment variable you set locks its setting. .env.example in the repository lists them all.

  1. Database: Postgres with pgvector from any provider.
  2. Storage: any S3-compatible bucket or the local file system.
  3. Sign-in: Google, GitHub, email links or local accounts.
  4. Transcription and questions: a managed provider, an OpenAI-compatible server on your network, or off.

Offline.

With OFFLINE_MODE=true, the instance serves the embedding model, ffmpeg and fonts itself, and the browser calls no provider directly. Point transcription and questions at a local Whisper server and Ollama to keep every request on your network.

Planned: a Compose file for air-gapped installs that adds the local Whisper server and Ollama.

On one computer.

askcalls local runs the whole product on your machine, with no server, no Docker and no account: an embedded Postgres, local file storage, and the app bound to 127.0.0.1. Your data stays encrypted on disk, and you still unlock it with a passkey, recovery code or passphrase.

Not available in this version: the command is registered in the CLI and lands with local mode.

Ask your calls anything.