Authentication
Flowgen supports user-level authentication for HTTP-facing tasks (webhooks, AI gateway, MCP server), and a separate interactive OIDC login for the web UI. Task-facing auth is configured once on each server section and shared across every HTTP-facing task on that server. When enabled, the resolved user identity is injected into the event metadata as event.meta.auth, where downstream tasks can read it for routing or audit.
User-level auth is separate from credentials_path — see Credentials for the distinction.
Providers
Three provider types, configured under http_server.auth (same shape works on mcp_server.auth and ai_gateway.auth):
| Provider | What it validates | When to use |
|---|---|---|
jwt | JWT signed with HMAC (HS256) or asymmetric keys (RS256/ES256). | You issue tokens yourself, or you have a JWKS endpoint. |
oidc | JWT issued by an OIDC provider, validated via discovery. | Auth0, Okta, Keycloak, Google, anyone with .well-known/openid-configuration. |
session | Opaque session token validated by an external HTTP endpoint. | You already have a session API and want flowgen to delegate validation. |
JWT provider
http_server:
enabled: true
auth:
type: jwt
secret: "your-hmac-secret" # for HS256 (mutually exclusive with jwks_url)
# jwks_url: "https://idp.example.com/.well-known/jwks.json" # for RS256/ES256
audience: "flowgen-prod" # optional: reject tokens with a different aud
issuer: "https://auth.example.com" # optional: reject tokens with a different iss
user_id_claim: "sub" # optional: defaults to "sub" Supply either secret (symmetric) or jwks_url (asymmetric), not both. With jwks_url, flowgen fetches the JWKS at startup and matches incoming tokens by their kid header.
The secret is held as a redacted value: it never appears in logs or the System view (config renders it as "***"). Token validation uses the real value.
OIDC provider
http_server:
enabled: true
auth:
type: oidc
issuer_url: "https://auth.example.com/realms/myapp"
audience: "flowgen-prod"
user_id_claim: "sub" Flowgen reads {issuer_url}/.well-known/openid-configuration at startup, extracts the JWKS endpoint, and validates incoming tokens against those keys. The provider rejects tokens whose iss claim does not match the discovered issuer.
Session provider
http_server:
enabled: true
auth:
type: session
validation_url: "https://auth.example.com/api/session/validate"
user_id_field: "user_id" For each incoming request, flowgen sends Authorization: Bearer <token> to validation_url. A 2xx response with a JSON body is treated as valid; the user_id_field is extracted from the response and becomes UserContext.user_id. Any other status returns 401 to the caller.
Per-task opt-in
Setting auth at the server level enables the provider but does not force every task to require a token. Each HTTP-facing task opts in individually:
- http_endpoint:
name: secure_endpoint
endpoint: /admin/events
method: POST
auth:
required: true When auth.required: true:
- Requests without an
Authorizationheader are rejected with 401. - Requests with an invalid token are rejected with 401.
- Requests with a valid token proceed;
UserContextis injected intoevent.meta.auth.
When auth.required: false (or auth is omitted on the task):
- Tokens are validated if present and injected into
event.meta. - Requests without a token still proceed (anonymous).
- This is useful for endpoints that personalise responses when the caller is logged in but stay public otherwise.
Reading user identity in scripts
Once event.meta.auth is set, downstream tasks can read it:
- script:
name: route_by_user
code: |
let user_id = event.meta.auth.user_id;
let role = event.meta.auth.claims.role;
if role == "admin" {
event.meta.lane = "priority";
}
event.data The shape of event.meta.auth:
{
"user_id": "[email protected]",
"claims": {
"sub": "[email protected]",
"email": "[email protected]",
"role": "admin",
"iss": "https://auth.example.com",
"exp": 1893456000
}
} user_id is the claim selected by user_id_claim (defaults to sub). claims is the full set of claims from the token, useful for fine-grained authorisation in scripts.
Composing with credentials_path
A webhook can require both a shared bearer secret (credentials_path) and a user-level JWT (auth.required: true). Both checks must pass. The shared secret is for service-to-service authentication (the caller proves they are an authorised service); the JWT identifies the end user.
- http_endpoint:
name: secure_endpoint
endpoint: /events
method: POST
credentials_path: /etc/flowgen/credentials/service-token.json
auth:
required: true The Authorization header in the incoming request carries the JWT. Service-to-service auth uses a separate header — typically X-Internal-Token — that the webhook compares against the credentials file. (Implementation details vary by deployment; check the webhook docs.)
Errors
| Status | Cause |
|---|---|
| 401 No token | Task has auth.required: true and no Authorization header was provided. |
| 401 Invalid token | Token signature, expiration, audience, or issuer check failed. |
| 401 Unknown key | JWKS-based JWT references a kid not present in the keyset. Refresh JWKS or verify the issuer. |
| 401 Session rejected | Session validation endpoint returned non-2xx. |
| 502 / 503 | Upstream JWKS endpoint or session validation service is unreachable. Retry by the caller. |
Worker logs include the specific reason at error level so operators can diagnose without leaking the token to the caller.
Web UI login
The token-validation providers above cover HTTP-facing tasks. The web UI (web.enabled: true) uses a separate, interactive OIDC login instead, configured under web.auth:
web:
enabled: true
port: 8080
path: "/flowgen"
auth:
issuer_url: "https://auth.example.com"
client_id: "flowgen-web"
client_secret: "your-oidc-client-secret"
# Must exactly match a redirect URI registered with the IdP, and
# include `path` above.
redirect_uri: "https://example.com/flowgen/auth/callback"
# extra_scopes: ["offline_access", "groups"]
# The provider's logout URL, ending its session too on sign-out.
# signout_redirect_url: "https://auth.example.com/v1/logout?post_logout_redirect_uri=https%3A%2F%2Fexample.com%2Fflowgen%2F"
# JSON file holding client_secret and cookie_secret.
# credentials_path: "/etc/flowgen/credentials/web.json"
cookie_secret: "a long random string, at least 32 bytes"
# cookie_secure: false # only for a plain-HTTP deployment; defaults to true Works with any standard-compliant OIDC provider — Okta, Zitadel, Auth0, or one that itself federates to an upstream identity provider (SSO broker setups look identical to flowgen, since discovery and token validation don’t change).
auth is optional; if omitted, the web UI is served unauthenticated.
Login cookies carry the Secure attribute by default, which browsers require HTTPS to send — set cookie_secure: false for a plain-HTTP deployment (local testing, or a proxy that already terminates TLS).
No server-side session store
Flowgen does not keep a session table. The browser’s cookie is the session: after login, it holds the identity provider’s ID and refresh tokens, encrypted with cookie_secret so the browser can carry it but never read or forge it. Every request re-validates the token; once it expires, flowgen transparently refreshes it against the identity provider. Concurrent requests to one instance share a single refresh, so a provider that rotates refresh tokens never sees one reused. With several replicas, keep a rotation grace period at the provider, since requests landing on different replicas refresh separately.
Refreshing needs a refresh token, which most providers issue only when the offline_access scope is requested and the application allows the refresh token grant. Without one, the session ends when the ID token expires and the browser goes back through the identity provider, returning to the page it was on. Signing out at the identity provider is what actually revokes access — flowgen has no session state of its own to invalidate.
A cookie secret is required whenever auth is set, from either web.cookie_secret or auth.credentials_path — flowgen refuses to start the web server without one, rather than falling back to an unauthenticated UI.
Keeping the secrets out of the config
client_secret and cookie_secret take three sources, in precedence order:
auth.credentials_path— a JSON file holding either or both keys:{ "client_secret": "the-oidc-client-secret", "cookie_secret": "a long random string, at least 32 bytes" }This is the same mounted-file convention the connectors use — see Credentials.
APP_WEB__AUTH__CLIENT_SECRETandAPP_WEB__COOKIE_SECRETin the environment. Every config key can be overridden this way:APP_prefixes the path and__separates each level.The inline values in the config file.
Signing out
Sign-out always clears flowgen’s session cookie. Set signout_redirect_url to the identity provider’s logout URL to end the provider’s session as well; flowgen appends id_token_hint so the provider knows which session to end.
Give the URL in full, including whatever return parameter the provider expects (commonly post_logout_redirect_uri, URL-encoded) — providers differ here, and not all of them advertise a logout endpoint in their discovery document. Whatever return URL it carries must be registered with the provider.
Without it, sign-out is local to flowgen: the provider still considers the browser signed in, so signing back in needs no credentials.
Routes
| Route | Purpose |
|---|---|
GET {path}/auth/login | Redirects to the identity provider. return_to names a web UI page to come back to after sign-in. 404 if auth isn’t configured. |
GET {path}/auth/callback | Identity provider redirect target; exchanges the code and sets the session cookie. When signing in fails, for example because the user is not assigned to the application, it redirects to the web UI’s login page, which says why; the provider’s own description is in the flowgen logs. |
GET {path}/auth/logout | Clears the session cookie, then redirects to signout_redirect_url or back to the UI. 404 if auth isn’t configured. |
GET {path}/auth/me | The signed-in user, 401 if not signed in, or 404 if auth isn’t configured. |
{path} is web.path (defaults to /).