NeuBird for Zscaler
Zscaler troubleshooting, straight to the root cause.
NeuBird is the AI SRE. It reads Zscaler ZIA, ZPA and ZDX alongside your application telemetry, cloud and Kubernetes, and tells your on-call engineer whether the problem is the app or the path, and which change caused it. Root cause arrives in under 5 minutes at 94% accuracy. NeuBird reads Zscaler and never changes its policy.
It's the path, not the app. payments-api is healthy: p99 142 ms, no errors, no deploy since yesterday. App Connector zpa-connector-usw2-02 went Unhealthy at 8:52 and adds 1,240 ms on the Cloud Path.
Cause: at 8:44 a Terraform change added ledger-sync.internal to application segment payments-internal, so ledger sync traffic now shares connector group acg-usw2.
Recommended: revert a41c9e2 in zpa-config. Pull request #412 is staged.
The problem
Why does the app look down when it isn't?In zero trust, the path is part of the app.
Every request to a private app crosses Zscaler Client Connector, a Zscaler Service Edge and an App Connector before it reaches a server. When users report that the app is down, the cause often sits on that path: an unhealthy App Connector, an access policy change, SSL inspection catching a new API endpoint, or one region's ZDX score falling. The app team sees healthy servers, the network team sees a healthy cloud, and the incident bounces between them.
- 1
One app goes Poor
The ZDX score for Payments Console falls from 81 to 31. Every other app holds steady.
- 2
Every region at once
San Jose, Frankfurt, London and Bengaluru drop together, so the cause sits after the users, not near them.
- 3
The path, hop by hop
The Cloud Path trace is fast to the ZPA Service Edge, then slow at one App Connector.
- 4
App or path?
The server itself answers in 38 ms. The 1,240 ms is spent in the path.
- Client ConnectorUser laptop · San Jose4 ms
- ISPHome broadband11 ms
- ZPA Service EdgeSan Jose9 ms
- App Connectorzpa-connector-usw2-021,240 ms
- payments-api.internalus-west-238 ms
Connect
What does NeuBird read from Zscaler?Five products, one OneAPI client.
NeuBird connects through the Zscaler OneAPI with an API client you create in ZIdentity, and it runs the official Zscaler MCP server with its write tools turned off. The roles on that client decide which products NeuBird can see, and NeuBird never changes Zscaler policy or configuration. Data is queried in place, with no bulk telemetry stored.
ZIA
Zscaler Internet Access
URL filtering, SSL inspection, cloud firewall and cloud app control policy, and when each rule last changed.
ZPA
Zscaler Private Access
Application segments, server groups, App Connector groups and access policies, with live App Connector status.
ZDX
Zscaler Digital Experience
ZDX scores by app, region and user, web and Cloud Path probe results, and device health.
Client Connector
Zscaler Client Connector
Device enrollment, forwarding profile and app profile state for the users an incident touches.
Z-Insights
Zscaler analytics
Analytics across Zscaler traffic, for trends that reach beyond a single incident.
Follow the path
Which change broke the path?Every recent change, on one timeline.
NeuBird follows an affected app through its application segment, server group and App Connector group to the connectors that carry it, and checks the health of each. Then it lines up what changed: ZPA and ZIA objects by the time they were modified, plus the commits in the repository that manages your Zscaler configuration. In this sample, a ZIA SSL inspection change at 7:32 AM is real but sits off the path. The 8:43 AM commit that put ledger sync traffic on the payments connectors is the one in the causal chain.
- 1
Follow the segment
Payments Console reaches users through application segment payments-internal and server group sg-payments-usw2.
- 2
Two connectors, one saturated
Connector group acg-usw2 has two App Connectors. One went Unhealthy at 8:52; the other is at 91% CPU.
- 3
Every recent change
NeuBird lists what changed in ZIA, ZPA and the zpa-config repository in the last two hours.
- 4
The one that matters
A commit at 8:43 added ledger-sync.internal to the payments segment. The ZIA change is not on this path.
payments-api.internalpayments-console.internalledger-sync.internal+ 8:44
zpa-connector-usw2-01Healthyzpa-connector-usw2-02HealthyUnhealthy- 7:32 AM
ZIASSL inspection rule
partner-apismodifiedNot on this path - 8:43 AM
zpa-configa41c9e2 merged: add ledger-sync to payments segmentIn the causal chain - 8:44 AM
ZPASegment
payments-internalmodified byterraform-zpaIn the causal chain - 8:52 AM
ZPA
zpa-connector-usw2-02status UnhealthyIn the causal chain
Resolve
How does NeuBird find the root cause?App and path, investigated together.
When the ZDX alert fires, NeuBird queries Zscaler beside your logs, metrics, traces, deploys, cloud and Kubernetes, in parallel across 15+ sources when an incident needs them. It rules the app out with evidence, names the causal chain with a citation for every step, and stages the fix in the change process you already use, all in under 5 minutes at 94% accuracy. Andrew Lee, the on-call engineer, approves it at a configurable human approval gate. The conclusion is saved to memory, with no bulk telemetry stored, so NeuBird recognizes the pattern the next time a segment change overloads a connector group.
- 1
The ZDX alert fires
Payments Console drops to Poor at 8:56 AM, and NeuBird starts before anyone opens a dashboard.
- 2
App and path, in parallel
NeuBird queries ZDX, ZPA and ZIA beside Datadog, Kubernetes and GitHub at the same time.
- 3
Root cause, cited
The app is ruled out with evidence. The causal chain runs from one commit to one connector.
- 4
Staged, then approved
NeuBird stages the revert. Andrew Lee, on call, approves it at the gate.
Kubernetesdeploy/payments-api · prod-usw2Running…12/12 pods Ready · 0 restartsThe path, not the app. A Terraform change routed ledger sync traffic through the payments App Connector group, and one of its two connectors saturated.
- 8:43
zpa-configcommit a41c9e2 merged[8] - 8:44
ledger-sync.internaladded to segmentpayments-internal[7] - 8:52Ledger sync traffic saturates
acg-usw2;-02goes Unhealthy[6] - 8:56ZDX score Poor; page fetch time 0.9 s → 6.8 s[1, 2]
- 1Revert a41c9e2 in zpa-configPull request #412 staged
- 2Give ledger-sync.internal its own segment and connector groupChange plan attached to the PR
- 3Add a third App Connector to acg-usw2Follow-up ticket for the network team
Prevent and Operate
What does NeuBird do before the page?And between incidents?
NeuBird watches ZDX scores against App Connector load and recent changes, and flags drift in one geography or a connector group running out of headroom 30 to 60 minutes before users file tickets. Between incidents it surfaces the hygiene that keeps zero trust reliable: connector groups with no redundancy, shadowed access policy rules and noisy experience alerts. Each finding carries the change NeuBird recommends, and the AI SRE acts only as far as your Suggest, Recommend and Act policy allows, set per action.
acg-euc1 CPU has climbed from 61% to 82%, 3 points every 6 minutes. At this rate it saturates about 10:41 AM. London and San Jose hold steady.acg-euc1 before the Frankfurt morning peak.- Connector group
acg-apse1runs one App Connector for 4 application segmentsSuggest: Add a second connector for redundancy - Access policy rule 22 is shadowed by rule 14 and has matched nothing in 30 daysSuggest: Remove it, or move it above rule 14
- ZDX alert
wifi-signal-weakfired 212 times last week and linked to 0 incidentsSuggest: Raise the threshold or send it to a daily digest - Application segment
legacy-billinghas had no user traffic in 90 daysSuggest: Review it for removal
Across the network path
Where else does NeuBird read the network?
Amazon CloudWatch
Already bringing Zscaler logs into CloudWatch with CloudWatch pipelines? NeuBird investigates them through one IAM role.
Palo Alto Networks Panorama
Firewall policy and device state from Panorama, investigated beside the rest of the network path.
Cisco SD-WAN
Tunnel, edge and policy state from the WAN, for incidents that start between sites.
All solutions
Every platform, tool and team NeuBird works with, across 50+ integrations.
FAQ
Common questions
What does NeuBird do for Zscaler?
NeuBird is the AI SRE. It reads Zscaler ZIA, ZPA, ZDX, Client Connector and Z-Insights alongside your application telemetry, cloud and Kubernetes, and answers the two questions every zero-trust incident starts with: is it the app or the path, and which change caused it. It isolates root cause in under 5 minutes at 94% accuracy.
How does NeuBird connect to Zscaler?
Through the Zscaler OneAPI. You create an API client in the ZIdentity Admin Portal, give it a role for each product NeuBird should see, such as ZIA, ZPA and ZDX, and enter its client ID, client secret and vanity domain in NeuBird. ZPA also needs your ZPA Customer ID. NeuBird needs outbound HTTPS to your ZIdentity login domain and to api.zsapi.net, and nothing is installed in your Zscaler tenant.
Can NeuBird change our Zscaler policy?
No. NeuBird runs the official Zscaler MCP server with its write tools turned off, so it reads configuration, status and experience data and never changes policy or configuration in Zscaler. The roles on the API client decide which products it can see. When NeuBird recommends a fix, it goes through the change process you already use, such as your Terraform pipeline, at configurable human approval gates.
Which Zscaler data does NeuBird use in an investigation?
ZDX scores, web and Cloud Path probe results and device health; ZPA application segments, server groups, App Connector groups, App Connector status and access policies; ZIA URL filtering, SSL inspection, firewall and cloud app control policy; Client Connector device and enrollment state; and Z-Insights analytics. NeuBird lines these up with your logs, metrics, traces, deploys and cloud changes, across 15+ sources in parallel.
Does NeuBird copy or store our Zscaler data?
No. NeuBird queries Zscaler through the OneAPI when an investigation needs it, with no bulk telemetry stored. Its memory holds conclusions, causal chains and approvals, never raw logs, metrics or traces.
How does NeuBird prevent Zscaler-related outages?
NeuBird watches ZDX scores against App Connector load and recent changes, and flags a drift, such as one geography degrading or a connector group running out of headroom, 30 to 60 minutes before users start filing tickets. Between incidents it surfaces hygiene work: App Connector groups with a single connector, overlapping or shadowed access policy rules, and noisy experience alerts.
We already send Zscaler logs to Amazon CloudWatch. Does NeuBird use them?
Yes. If CloudWatch pipelines bring your Zscaler logs into CloudWatch, NeuBird investigates them through the same IAM role as the rest of your AWS telemetry. The Zscaler connection adds what logs do not carry: live App Connector status, ZDX scores and path traces, and the current policy configuration.
NeuBird for Zscaler
App or path, answered in under 5 minutes.
NeuBird, the AI SRE, reads Zscaler beside the rest of your stack, prevents the next outage and remembers every fix, with no bulk telemetry stored.
