Nojoin Administration Guide

This guide is for Owners and Admins running a self-hosted Nojoin installation.

Roles

Nojoin supports three primary roles:

Owner: Full system access, including server configuration and user management.
Admin: Can manage users and invitations, but cannot take the highest-privilege owner actions.
User: Standard access to personal recordings, settings, and connected services.

Additional privilege guardrails apply around Owner creation and superuser-only operations.

Invitations

Registration is invite-gated.

Admins can:

Create invitation links.
Choose the invited role between user and admin.
Set expiry and usage limits.
Revoke invites while retaining the historical record.
Delete revoked invites permanently if desired.

Users who register through an invite choose their own password during sign-up and are not forced through an immediate password-rotation flow.

Manual User Provisioning and Password Rotation

New and rotated passwords are enforced server-side.

Passwords must be at least 8 characters long.
Passwords made entirely of whitespace are rejected.
Existing password hashes are grandfathered until the next password change or admin reset.

When an Admin or Owner creates a user manually:

The user receives a temporary password.
The user must choose a new password before the rest of the application becomes available.
While force_password_change is active, Nojoin only allows self-profile access, password update, and logout.

The same restriction also applies when a superuser resets another user’s password through the privileged user-management flow.

Administration Settings Areas

Calendar

Use Settings > Administration > Calendar providers to save installation-wide Google and Microsoft OAuth credentials.

Read CALENDAR.md for the full provider registration and tenant guidance.

AI and Models

Use Settings > AI for installation-wide provider defaults, model operations, and Ollama configuration.

[!IMPORTANT] For security, LLM provider API keys and Hugging Face tokens are server-side environment-only variables and must be configured via environment variables (e.g., .env) and the container restarted, rather than through UI settings fields.

Admin-only sections let you:

Choose the default LLM provider.
Configure the Ollama API URL and context window. The context window is sent to Ollama as num_ctx for full-context meeting prompts; if Ollama still reports a length stop, Nojoin surfaces that as a chat error instead of saving a truncated answer.
Configure a secondary LLM provider for fallback. When the primary provider fails, the system automatically retries with the secondary provider. The secondary provider has its own independent configuration (provider, model, API key, Ollama URL) set through SECONDARY_ prefixed environment variables.
View installed Whisper models.
Remove local model cache entries. Required default models are prepared automatically, and repo-bundled model assets remain read-only in the UI.

Each user can also configure Language preferences in Settings > AI. Transcription defaults to automatic language detection, while generated meeting titles and notes default to English. Whisper supports forced language selection, Canary supports its listed language set, and Parakeet remains automatic-only. The selected effective transcription language is part of the ASR reuse key, so cached/live transcript work is not reused across incompatible language settings.

Notes-language choices include British and American English, the transcription language, listed languages, and a validated custom language/style instruction. These choices localize generated content while preserving machine-readable JSON keys and speaker labels. Existing saved transcripts and notes are not translated in place; users must reprocess or run Generate Notes after changing the relevant preference.

Backup and Restore

Use Settings > Administration > Backup and restore for export and restore operations.

Read BACKUP_RESTORE.md before relying on it operationally, especially because backup archives can contain restorable calendar credentials and connected-account tokens.

System

Use Settings > Administration > System operations for operational controls such as:

Restarting the stack.
Viewing live logs.
Filtering merged or per-service log output.
Downloading log output for investigation.

Updates

Use Settings > Updates to see:

The installed server version from the current API build.
The latest stable published release.
Release history and release notes.

Operational Notes

Back up the installation before upgrading.
Review release notes for browser capture, auth, and upload lifecycle changes.
After live-pipeline upgrades, use the recording page’s waveform, Meeting Edge status, and overall recording progress before treating a meeting as stuck.
For remote deployments, configure a trusted public origin with WEB_APP_URL.
Treat backup archives as sensitive material.

Browser Capture Support

Shared-audio live recording requires Chrome on Windows, Linux, or macOS, or another supported desktop Chromium browser. Other Chromium-family browsers on macOS are best-effort.
Chrome on Android and iOS can start microphone-only live recordings.
Firefox, Safari, and other mobile browsers can review and administer Nojoin but cannot start live capture.
Tab sharing with audio enabled is the recommended support path for browser-based meetings.
If local microphone audio is missing, ask the user to grant microphone permission and review Settings > Capture.
If remote participant audio is missing, ask the user to start again and enable shared audio in the browser picker.
If a mobile Chrome recording is missing remote participants, confirm the user expected microphone-only capture and that the phone microphone could hear the meeting audio.
If a user has a paused recording, they must resume or discard it before starting another capture.
Review backend and worker logs for segment upload, transcode, live transcription, finalize, or discard failures.
Browser-live segment numbering starts at 0; upload or finalize support cases should confirm the sequence is contiguous.
The worker keeps browser-live audio as 16 kHz, two-channel WAV after transcode. Channel 0 is shared/system audio when available and channel 1 is microphone audio.

Paused recordings are retained indefinitely and are not cleaned up automatically. This protects uploaded meeting data and prevents overlapping segment streams for the same user.

Read CAPTURE.md for the support matrix, browser picker behaviour, and troubleshooting steps.