Skip to main content

ForwardAuth & Project Access Control

muvee supports per-project authentication for deployed applications. When enabled, any visitor to {project}.domain.com is first redirected to sign in via whichever OAuth provider you have configured (Google, Feishu, WeCom, DingTalk, etc.).

How It Works​

Browser → Traefik → ForwardAuth sidecar → OAuth provider
                         ↓
                    Validates JWT session cookie
                         ↓
                    Checks email domain against
                    project.auth_allowed_domains
                         ↓
                    200 OK → Traefik forwards request
                    403 → access denied

Enabling Auth on a Project​

In the Project Detail → Auth tab:

  1. Toggle Require Auth to Enabled
  2. Optionally set Allowed Email Domains (comma-separated, e.g. company.com,partner.org)
  3. Optionally set Auth Bypass Paths — one path per line. Requests matching these paths will skip authentication. Use * suffix for prefix matching (e.g. /api/public/*).
  4. Save and redeploy

When the container starts, the deploy agent attaches these Traefik labels:

traefik.http.middlewares.{proj}-auth.forwardauth.address=http://muvee-authservice:4181/verify?project={id}&domains=company.com
traefik.http.middlewares.{proj}-auth.forwardauth.authResponseHeaders=X-Forwarded-User,X-Forwarded-User-Name,X-Forwarded-User-Avatar,X-Forwarded-User-Provider
traefik.http.routers.{proj}.middlewares={proj}-auth

Session Flow​

  1. User visits {project}.domain.com
  2. Traefik calls ForwardAuth sidecar with the request headers
  3. Sidecar checks muvee_fwd_session cookie (JWT)
  4. If missing/expired: redirect to the configured OAuth provider (/_oauth callback on BASE_DOMAIN)
  5. After login: set JWT cookie (domain-wide, shared across all *.BASE_DOMAIN subdomains), redirect back to original URL
  6. On subsequent requests: validate JWT, check email domain, return 200
How the OAuth callback is routed

The /_oauth path on BASE_DOMAIN is routed by Traefik directly to muvee-authservice (not to the main web UI). This is configured via Traefik labels on the muvee-authservice container in docker-compose.yml. Because Traefik gives higher priority to the more specific Host + Path rule, BASE_DOMAIN/_oauth is correctly handled by the auth sidecar while all other BASE_DOMAIN paths continue to reach muvee-server.

Auth Bypass Paths​

When auth is enabled on a project you can exempt specific paths from authentication. This is useful for health checks, public APIs, or webhook endpoints that need to be reachable without a session.

Configure bypass paths in the Auth tab (one path per line) or via the CLI:

muveectl projects update PROJECT_ID --auth-bypass-paths "/health
/api/public/*"
PatternMatches
/healthExact path /health only
/api/public/*Any path starting with /api/public/

Each bypass path creates a higher-priority Traefik router that routes directly to the service without the ForwardAuth middleware.

Public Projects​

If Require Auth is disabled on a project, no ForwardAuth middleware is attached — the project is publicly accessible.

Per-Project Access Control (access_mode)​

When Require Auth is enabled, every project also has an access_mode that decides which signed-in users are actually allowed to reach the deployed service. It is independent from the email-domain check above and runs after it.

access_modeWho can reach the service
public (default)Any signed-in muvee user.
privateThe project owner, system admins, and users explicitly listed in the project's allow-list.

The decision is made by muvee-server's /api/internal/access/check endpoint (called by the auth sidecar on every request) and recorded in the project's Users tab.

Managing access from the UI​

Open the project's Users tab to manage all of this in one place:

  • Service access — toggle between Public and Private. Owners and system admins are always allowed regardless of mode and never need to be added to the allow-list.
  • Allowed users — the explicit allow-list consulted when access is Private. Users have to sign in to muvee at least once before they can be invited (the lookup is by email).
  • Recent visitors — unique users who have actually reached the deployed service, with a per-user visit count and last-seen timestamp. Useful for both modes:
    • For Public projects: see who's actually using your service.
    • For Private projects: each row gets a one-click Allow button that adds that user to the allow-list without having to re-type their email.
  • Pending requests — when a denied user submits an access request (see below), it appears here with the reason they gave. Approve to add them to the allow-list immediately; deny to reject.

The sidebar Projects link shows a small badge with the total number of pending access requests across every project you own. It refreshes every minute.

Request-access flow for denied users​

When a user is denied by a Private project, the auth sidecar redirects them (HTTP 302) to https://BASE_DOMAIN/request-access?project={id} instead of returning a bare 403. There they:

  1. Sign in to muvee if they haven't already.
  2. See the project they were trying to reach.
  3. Optionally enter a short reason (≤ 1000 chars).
  4. Submit. The request lands in the project's Users → Pending requests panel.
  5. Once the owner approves, the next visit succeeds without further prompting.

If the same user submits multiple times before a decision, the existing pending row is reused (not duplicated). After a decision the row stays around for history; a denied user can re-request later by submitting again. Approval is permanent in the sense that the user is added to project_access_users immediately; revoking access requires the owner to remove them from Allowed users.

How visit recording works​

Visit counts are written through an in-memory batch worker, not on the request hot path:

  • Every successful access check pushes one event onto a 1024-slot channel (non-blocking — events are dropped with a warn log if the buffer ever saturates).
  • A background goroutine flushes the channel every 5 seconds, or whenever 200 events have accumulated, into a single multi-row UPSERT against project_visits.
  • Same-user events that arrive in the same batch are deduplicated; their counts are summed and last_seen_at is set to GREATEST(existing, new) so out-of-order arrivals never move the timestamp backwards.
  • On graceful shutdown the worker drains the channel and does one final flush. A hard kill can lose up to ≤ 5 seconds of visit counters.

This keeps the ForwardAuth path one DB round-trip (the access check itself) instead of two.

For Service Developers​

If you are developing a service deployed on muvee and want to know how to read user identity, implement logout, or support CLI access, see the Service Auth Integration guide.