NeuBird and Cribl
Cribl root cause analysis, from data your pipelines already cleaned.
NeuBird is the AI SRE, and it works with Cribl. It runs Cribl Search queries against the datasets you already route to Cribl Lake, joins the firewall, Kubernetes and application logs your pipelines have given one shape, and posts the root cause in under 5 minutes, at 94% accuracy. Your telemetry stays in your Cribl tier, and NeuBird stores no bulk telemetry.
dataset="prod_app_logs" service="checkout-svc" | where status >= 500 | summarize count() by error, upstreamRunning…1,842 × upstream connect timeout · payments-api:8443dataset="prod_firewall" dest_port=8443 action="deny" | summarize count() by rule, src_ipRunning…1,796 denies · rule deny-legacy-8443 · 10.42.0.0/16A 14:01 commit on pan-fw-edge-01 added rule deny-legacy-8443, which blocks checkout-svc pods from payments-api.
Cribl Stream
How does Cribl Stream shape telemetry?Clean, consistent events before they land.
Cribl Stream sits between your sources and your destinations. Its pipelines run Functions on every event in flight: Parser extracts fields, Eval normalizes names, Mask hashes sensitive values, Suppress removes duplicates and Drop discards the noise. Routes then send the result to Cribl Lake and any other destination you run, and Packs let teams reuse proven pipelines, including Packs that output common schemas such as OCSF.
- 1
Raw events arrive
Firewall syslog, Kubernetes logs and application JSON, each in its own format and field names.
- 2
A pipeline shapes them
Cribl Stream Functions parse, normalize, mask and trim every event in flight.
- 3
Routed to Cribl Lake
Routes send the clean events to Cribl Lake datasets and to any other destination you run.
- 4
One shape across sources
The same connection reads src_ip and dest_port in the firewall log and the application log alike.
prod_firewallprod_k8s_logsprod_app_logs<14>Nov 20 14:02:11 pan-fw-edge-01 1,2026/11/20 14:02:11,TRAFFIC,end,10.42.7.19,10.80.3.24,deny-legacy-8443,tcp,52214,8443,deny{"ts":"2026-11-20T14:02:11.402Z","svc":"checkout-svc","msg":"upstream connect timeout","peer":"10.80.3.24:8443","client":"10.42.7.19","user":"a.lee@example.com"}Better together
Why does Cribl data make a better AI SRE?Cribl shapes and keeps it. NeuBird reasons over it.
Cribl Lake keeps that clean data in open formats, with retention you set for each dataset, and Cribl Search queries it in place, along with object storage and Cribl Edge. NeuBird connects through Cribl Search, so the work Cribl already did on your data carries into every investigation.
One shape, one investigation
Normalized fields let NeuBird join firewall, Kubernetes and application logs on src_ip, dest_port and service, with no query translation for each source.
Less noise, fewer tokens
Events arrive parsed, deduplicated and trimmed, so NeuBird asks fewer, smaller questions. Model context stays lean and token use stays efficient.
Your data stays in Cribl
NeuBird queries Cribl Search in place and stores no bulk telemetry. Retention, access and cost stay where your Cribl team set them.
Resolve · when systems break
How does NeuBird find root cause in Cribl data?Targeted Cribl Search queries, joined on shared fields.
When an alert fires, NeuBird plans its questions and runs Cribl Search queries against the datasets that can answer them, as part of the 15+ sources it analyzes in parallel. In this sample investigation, checkout-svc starts returning 5xx errors. The application logs point at payments-api, the Kubernetes logs clear its pods, and the firewall dataset shows a new rule denying the traffic. One pipeline normalized all three, so NeuBird lines them up by src_ip and dest_port and posts the root cause in under 5 minutes.
- 1
The alert fires
checkout-svc returns 5xx errors. NeuBird starts before anyone opens a dashboard.
- 2
Search each dataset
Targeted Cribl Search queries against application, Kubernetes and firewall datasets.
- 3
Join on one field set
The same src_ip and dest_port line the firewall denies up with the application timeouts.
- 4
Root cause in under 5 minutes
A cited causal chain, a staged fix and an approval gate for the on-call engineer.
dataset="prod_app_logs" service="checkout-svc"
| where status >= 500 | summarize count() by error, upstreamRunning…1,842 × upstream connect timeout · payments-api:8443dataset="prod_k8s_logs" namespace="payments"
| where reason in ("OOMKilled", "BackOff", "Unhealthy") | summarize count() by podRunning…0 events · all payments-api pods readydataset="prod_firewall" dest_port=8443 action="deny"
| timestats span=1m count() by rule, src_ipRunning…1,796 denies since 14:02 · rule deny-legacy-8443dataset="prod_firewall" type="CONFIG" host="pan-fw-edge-01"
| project _time, cmd, pathRunning…14:01 commit · security rule deny-legacy-8443 addedA 14:01 commit on pan-fw-edge-01 added rule deny-legacy-8443. It blocks checkout-svc pods (10.42.0.0/16) from payments-api on port 8443.
deny-legacy-8443 on pan-fw-edge-01.Approval gate · awaiting Andrew LeeApproved by Andrew Lee · 14:08Prevent · before the page
Can NeuBird catch a broken Cribl source before dashboards go blank?Drift flagged 30 to 60 minutes early.
A source that stops sending looks like a quiet night until a dashboard goes blank or a no-data alert finally trips. NeuBird watches event volume per source through Cribl Search and checks a sudden drop against the source's peers and Cribl's own internal logs. In this sample, a certificate rotation cuts off one firewall's syslog at 10:12. NeuBird flags it at 10:19, with the cause, 53 minutes before the downstream alert would fire.
- 1
Watch volume per source
NeuBird tracks event counts per host and dataset through Cribl Search.
- 2
A source goes quiet
pan-fw-edge-02 stops sending at 10:12 while its two peers hold steady.
- 3
Find why
Cribl's own logs show the syslog source rejecting the firewall's new TLS certificate.
- 4
Flagged 53 minutes early
The downstream no-data alert would not trip until 11:12. NeuBird flags it at 10:19.
dataset="prod_firewall" | timestats span=1m count() by hostdataset="cribl_internal_logs" | where message has "certificate" | summarize count() by channelRunning…312 × certificate verify failed · input:syslog_panChange CHG0031 at 10:11 rotated the syslog TLS certificate on pan-fw-edge-02 to one your Cribl syslog source does not trust. Its traffic and threat logs have not reached prod_firewall since.
Operate · between incidents
Which Cribl sources earn their volume?Recommendations your pipeline owners can apply.
Between incidents, NeuBird compares each source's share of volume with how often investigations cite it. High-volume sources that never show up as evidence become recommendations for Cribl Functions such as Sampling, Aggregations or Drop, and the logs that carry root causes stay at full fidelity. Your team applies each change in Cribl Stream, where your pipelines already live.
- 1
Volume by dataset
NeuBird measures what each source contributes, through Cribl Search.
- 2
Against what investigations used
It counts how often each source was cited in the last 41 investigations.
- 3
Recommend a Cribl Function
Sampling, Aggregations or Drop for the noise; full fidelity for the evidence.
- 4
Your team makes the change
Recommendations go to the pipeline owner. NeuBird never edits a pipeline.
prod_k8s_logs · level=debug38%0 of 41Sample 10:1 with Samplingprod_firewall · action=allow27%3 of 41Summarize per minute with Aggregationsprod_app_logs · path=/healthz11%0 of 41Drop in Streamprod_app_logs · level=error6%29 of 41Keep at full fidelityprod_firewall · action=deny, CONFIG4%17 of 41Keep at full fidelityThe three noisiest sources make up 76% of volume and appear in 3 of 41 investigations. Recommendations go to the pipeline owner, who applies them in Cribl Stream.
Policy: Recommend · awaiting Andrew LeeConnect
How do you connect NeuBird to Cribl?One OAuth credential, scoped in Cribl.
Create an API credential in Cribl with access to Cribl Search and the datasets you want searched, then enter your Cribl.Cloud workspace URL or on-prem leader URL, the Client ID and the Client Secret in NeuBird. NeuBird is deployable in under one hour. Every action it takes runs through configurable human approval gates, with Suggest, Recommend or Act set per action, and every verified root cause is remembered as a conclusion, with no bulk telemetry stored.
- 1
Create a credential in Cribl
Manage, Organization, API Credentials. Give it Cribl Search and the datasets NeuBird should read.
- 2
Enter it in NeuBird
Your Cribl.Cloud workspace URL or on-prem leader URL, the Client ID and the Client Secret.
- 3
Scope stays in Cribl
NeuBird searches only the datasets the credential can read.
- 4
Connected
NeuBird stores the secret encrypted and mints OAuth tokens against your tenant on demand.
cribl-prodhttps://main-<orgId>.cribl.cloudnb-search-7f3c91ad••••••••••••••••••••prod_app_logsCribl Lake · application logsprod_k8s_logsCribl Lake · Kubernetes logs and eventsprod_firewallCribl Lake · firewall traffic, threat, configcribl_internal_logsCribl's own logsPermissions are set on the credential in Cribl. Datasets it cannot read are never queried.
Cribl often feeds the rest of your stack, and NeuBird connects there too, across 50+ integrations including Splunk, Datadog, Amazon CloudWatch and Kubernetes. See every NeuBird solution.
FAQ
Common questions
What does NeuBird do with Cribl?
NeuBird is the AI SRE. It connects to your Cribl.Cloud workspace or on-prem Cribl leader and runs Cribl Search queries against the datasets you already route, such as Cribl Lake datasets. It uses that evidence to isolate root cause in under 5 minutes at 94% accuracy, catch drift 30 to 60 minutes early, and run the hygiene between incidents.
How does NeuBird connect to Cribl?
With an OAuth API credential. In Cribl, an admin creates a credential under Manage, Organization, API Credentials and gives it access to Cribl Search and the datasets NeuBird should query. In NeuBird you enter your Cribl base URL, the Client ID and the Client Secret. NeuBird stores the secret encrypted and mints access tokens against your tenant on demand.
Does NeuBird change our Cribl Stream pipelines?
No. NeuBird searches the data your pipelines produce through Cribl Search. It does not configure Routes, Pipelines, Packs or Destinations and does not write to Cribl. When NeuBird recommends a change, such as sampling a noisy source, your team makes it in Cribl Stream.
Why does Cribl data make NeuBird investigations faster?
Cribl Stream pipelines parse, normalize and enrich events before they land, so firewall, Kubernetes and application logs share field names. NeuBird can join them in one investigation without translating a query for each source, and less noise means fewer, smaller queries and efficient token use.
Does NeuBird copy our telemetry out of Cribl Lake?
No. NeuBird queries Cribl Search in place and stores no bulk telemetry. Its memory holds conclusions, causal chains and approvals, never raw logs, metrics or traces, so your data stays in your Cribl tier under your retention policies.
Can NeuBird act on what it finds?
Yes, through configurable human approval gates. You set the Suggest, Recommend or Act policy per action. A high-risk change, such as reverting a firewall rule, waits for one-click approval, while a routine fix that matches pre-approved rules can run on its own. Every action is recorded in the audit trail.
NeuBird and Cribl
Cribl shapes the data. NeuBird, the AI SRE, finds the cause.
See NeuBird run Cribl Search against your datasets and post a cited root cause in under 5 minutes.
