Sending security events to your SIEM¶
An installation in your own infrastructure can send its security events — sign-ins, changes of members and roles, API keys — to your SIEM: MaxPatrol SIEM, KUMA, RuSIEM, QRadar, ArcSight, Splunk, Elastic or any other system that receives syslog. This page is for the engineers who install the platform and the security team that receives the events.
What is sent¶
| Event | When it happens | What it names |
|---|---|---|
user.login |
A person signs in with a password, a social account or your SSO, or is refused | Who, from which address, the reason for a refusal, the SSO provider and the person's id there |
user.logout |
A person signs out | Who |
user.signup |
An account is created, or the attempt is refused | Who, the reason for a refusal |
user.delete |
A person deletes their account, or is refused | Who, the reason for a refusal |
member.invite |
Someone is invited to the organization | Who invited whom |
member.invite_revoke |
An invitation is withdrawn | Who withdrew which invitation |
member.join |
A person joins the organization: accepts an invitation or is added directly | Who joined and how |
member.remove |
A member is removed from the organization | Who removed whom |
member.role_change |
A member's role changes | Who changed it, whose, from which role to which |
api_key.create |
An API key of an agent is issued | Who issued it, the agent, the key's name and first characters |
api_key.revoke |
An API key is deleted | Who deleted it, the agent, the key |
Every event also carries the time, the result (success or failure), the client address and browser, the organization, and a unique event id.
With your SSO (for example Keycloak), the sign-in event carries the person's id at your identity provider, so the SIEM can join it with the provider's own log. Second-factor checks happen at your provider and are logged there.
How events are delivered¶
An event is recorded together with the action: if the action did not happen, there is no event; if it happened, the event is kept until your SIEM has received it. A separate service, the SIEM forwarder, sends the recorded events in the order they happened.
- When the SIEM is unreachable, nothing is lost: the forwarder retries with growing pauses (up to a minute) and sends everything that piled up once the receiver is back.
- An event may arrive twice after a failure in the middle of a delivery. Each event has a unique id (
externalIdin CEF,event.idin JSON); use it to drop repeats. - Events are kept for 90 days, delivered or not. An event your SIEM has not received in 90 days is deleted.
Formats¶
Syslog with CEF¶
The main option: syslog (RFC 5424) over TCP, with or without TLS, one event per line, with the event in Common Event Format (CEF) inside. Syslog facility is 13 (log audit); a refused attempt has severity warning, the rest notice.
<108>1 2026-11-13T09:41:07.312Z agent-platform security-audit - user_login - CEF:0|Vendor|Agent Platform|1.0|user.login|user login|6|rt=1794562867312 externalId=5d7a304e-3fd1-4159-bf1e-e17b896b7071 act=user.login cat=authentication outcome=failure reason=invalid_credentials [email protected] src=10.20.0.15 requestClientApplication=Mozilla/5.0 …
Vendor is the name of your installation's brand. The fields:
| CEF field | Meaning |
|---|---|
rt |
Time of the event, milliseconds since 1970 (UTC) |
externalId |
Unique event id |
act |
Event, from the table above |
cat |
authentication for sign-in and sign-out, iam for the rest |
outcome |
success or failure |
reason |
Why the attempt was refused, for example invalid_credentials, account_disabled |
suid, suser |
Who acted: id and email |
src |
Client IP address |
requestClientApplication |
Client's browser or app |
duid, duser |
Whom or what the action was applied to: id and name or email |
cs1 (organizationId) |
Organization |
cs2 (targetType) |
Kind of object: user, member, invite, api_key |
cs3 (traceId) |
Trace id, to find the request in the platform's traces |
cs4 (requestId) |
Request id |
cs5 (labels) |
Details as key:value;…: the SSO provider and the person's id there (provider, idp_subject), the roles before and after (role_from, role_to), the key's agent and first characters (agent_id, key_prefix) |
Fields with no value are left out.
JSON lines¶
If you collect container logs yourself (Fluent Bit, Vector, Filebeat), the forwarder can write each event as one JSON line to its log instead of, or in addition to, syslog. The fields follow the Elastic Common Schema (@timestamp, event.action, event.outcome, user.email, source.ip, …) and every line has "log": {"type": "audit"}, so one filter separates the events from everything else. The forwarder writes nothing else to that stream.
Turning it on¶
In the installation's values file, fill in the siem block and enable it:
siem:
enabled: true
syslog:
host: siem-collector.example.local # your syslog receiver
port: 6514 # 514 is customary for plain TCP
tls: true
caSecretName: siem-ca # CA your receiver's certificate is issued by
hostname: agent-platform # how the SIEM names this source
stdout: false # true: also write JSON lines to the log
If the receiver's certificate is issued by your own certificate authority, put the authority's certificate into a Kubernetes secret first and name it in caSecretName:
Apply the values as usual. The forwarder starts as a separate service, siem-forwarder, and begins with the events recorded so far.
Receiving in your SIEM¶
On the SIEM side you need a TCP syslog receiver on the port above, with TLS if you enabled it, that parses CEF:
- KUMA: a collector with a TCP connector and the CEF normalizer.
- MaxPatrol SIEM: a syslog data source over TCP with CEF normalization.
- QRadar, ArcSight, Splunk, Elastic: their standard TCP syslog input with CEF parsing.
The exact screens depend on your SIEM's version; your SIEM administrator sets this up as for any other CEF source. Use externalId to drop repeated events.
Checking that events arrive¶
- Sign in to the platform with a wrong password. Within a few seconds a
user.loginevent withoutcome=failureshould appear in the SIEM. - If it does not, look at the forwarder's log:
kubectl -n platform logs deploy/<release>-siem-forwarder. A receiver that cannot be reached, or a certificate that is not trusted, is reported there on every retry. - If the cluster runs Prometheus, the forwarder reports
security_audit_pending_events(events not yet delivered) andsecurity_audit_oldest_pending_age_seconds(how long the oldest of them has waited). TheSiemDeliveryStalledalert fires when an event has waited longer than 15 minutes.