Your first investigation
Choose the system you want to ask about, then follow eight steps from installing Verbiqo to a first answer you can check. Allow about half an hour, plus any time waiting for an administrator.
Choose your source
Each step says where to go, what you supply, what success looks like, and whether you need an administrator. Open If it fails under a step only when you need it.
Before you start
- A Mac with macOS 13 or later. On Windows the beta is given out on request, and the menu is by the clock instead of in the menu bar.
- An AI model you are allowed to send your systems’ metadata to: where your model comes from.
- Someone who can create a read-only account on the system, if that is not you.
1. Install Verbiqo
- Where
- The download section of verbiqo.app. Open the downloaded file, drag Verbiqo to Applications, then open it.
- You supply
- The right download: Apple Silicon for M-series Macs, Intel for older ones.
- Success
- The Verbiqo icon appears in the menu bar. There is no main window.
- Administrator
- Only if your Mac does not let you install apps.
If it fails
If macOS will not open it, check it came from verbiqo.app: the app is signed and notarized by Apple. Install details.
2. Set up a model that can investigate
- Where
- Menu bar icon › Settings… › AI provider, then Test & Load Models, Model and Save.
- You supply
- Your organization’s endpoint, an approved cloud API, or a model on this computer (which to choose), and Context window 32768, Timeout (ms) 300000, Temperature 0.2.
- Success
- Test & Load Models lists your models. Then, in Terminal,
/Applications/Verbiqo.app/Contents/Helpers/verbiqo model-checkreports Fit for Verbiqo. It uses built-in test data and connects to nothing but the model. - Administrator
- For your organization’s endpoint details, if you use one.
If it fails
Not fit as configured lists what to change. The models that passed are in Tested models. An unavailable model is usually the endpoint, the key or a certificate: Troubleshooting.
3. Get a read-only account
PostgreSQL
- What it is
- A login role that is not a superuser, granted
pg_monitor, and apg_hba.confline on every node that lets it connect. It reads activity, statistics and settings, and nothing in your tables. - Where
- The commands are in Set up PostgreSQL.
- Success
- The account exists and can sign in, with only those rights.
- Administrator
- Usually yes: a database administrator creates the role and edits
pg_hba.conf.
MySQL
- What it is
- A monitoring account with
PROCESSandREPLICATION CLIENT, and reads inperformance_schemaandsys. Nothing in your schemas. - Where
- The commands are in Set up MySQL.
- Success
- The account exists and can sign in, with only those rights.
- Administrator
- Usually yes: someone who can create accounts and grant privileges.
SQL Server
- What it is
- A login with
VIEW SERVER STATEandVIEW ANY DEFINITION, andSHOWPLANfor its user inmaster. - Where
- The commands are in Set up SQL Server.
- Success
- The account exists and can sign in, with only those rights.
- Administrator
- Usually yes: a database administrator creates the login.
Redis
- What it is
- An ACL user with no keys and no channels, and a few server-level read commands. Create it on every server, because ACL users are not copied.
- Where
- The commands are in Set up Redis.
- Success
- The account exists and can sign in, with only those rights.
- Administrator
- Usually yes: whoever administers the Redis servers.
Elasticsearch or OpenSearch
- What it is
- Elasticsearch: a monitoring role with an API key made from it. OpenSearch: a monitoring role mapped to a user or directory account. Cluster questions read health, statistics and index metadata rather than documents; security-log access is separate.
- Where
- The commands are under Elasticsearch role and OpenSearch role.
- Success
- The account exists and can sign in, with only those rights.
- Administrator
- Usually yes: someone who can create the monitoring role and credential.
Kubernetes
- What it is
- A ClusterRole that can only
getandlist, bound to a service account. Reading pod logs is optional. An administrator’s kubeconfig is refused. - Where
- The commands are in Set up Kubernetes.
- Success
- The account exists and can sign in, with only those rights.
- Administrator
- Yes: creating the role and binding needs a cluster administrator.
Prometheus or Thanos
- What it is
- Prometheus has no roles. If a token or proxy sits in front, it must allow reads of the query, series, label values, metadata and build information endpoints.
- Where
- The commands are in Set up Prometheus or Thanos.
- Success
- The account exists and can sign in, with only those rights.
- Administrator
- Only if a token or proxy controls access.
Splunk
- What it is
- A role with the
searchcapability, your log indexes insrchIndexesAllowed, and read on the app it searches in. - Where
- The commands are in Set up Splunk.
- Success
- The account exists and can sign in, with only those rights.
- Administrator
- Usually yes: a Splunk administrator creates the role.
4. Add the connection and test it
First, in Settings, tick Let Verbiqo read databases and other systems.
PostgreSQL
- Where
- Menu bar icon › Extensions… › PostgreSQL.
- You supply
- A name you will type later (for example orders), host, port 5432, database, user
verbiqo, the password, and TLS: Verify full with your CA certificate, or Require until you have it. - Success
- Test Connection shows Connected., then the version, primary or standby, sessions in use and the TLS in use. Press Save Connection.
- Administrator
- Only for network access or
pg_hba.conf.
If it fails
The test says what to fix: a password, a host that cannot be reached, a certificate, or an account refused because it could write or administer. Common errors · Connection problems.
MySQL
- Where
- Menu bar icon › Extensions… › MySQL.
- You supply
- A name (for example shop), host, port 3306, user, password, and TLS: Verify identity with your CA certificate, or Required for MySQL’s own self-signed certificate.
- Success
- Test Connection shows Connected., then the product, version, role (source, replica or group member) and the TLS in use. Press Save Connection.
- Administrator
- Only for network access.
If it fails
The test says what to fix: a password, a host that cannot be reached, a certificate, or an account refused because it could write or administer. Common errors · Connection problems.
SQL Server
- Where
- Menu bar icon › Extensions… › SQL Server.
- You supply
- A name (for example sales), host, port 1433 or the instance name, login, password, and encryption: Mandatory with Trust server certificate for the server’s own certificate, or Mandatory with your CA certificate.
- Success
- Test Connection shows Connected., then the product, version, edition, sessions and the encryption in use. Press Save Connection.
- Administrator
- Only for network access.
If it fails
The test says what to fix: a password, a host that cannot be reached, a certificate, or an account refused because it could write or administer. Common errors · Connection problems.
Redis
- Where
- Menu bar icon › Extensions… › Redis.
- You supply
- A name (for example cache), host, port 6379, user
verbiqo_monitor, password, and TLS: Disable for the plain port, or Verify full with the CA on the TLS port. - Success
- Test Connection shows Connected., then the product, version, role, clients, keys and the TLS in use. Press Save Connection.
- Administrator
- Only for network access.
If it fails
The test says what to fix: a password, a host that cannot be reached, a certificate, or an account refused because it could write or administer. Common errors · Connection problems.
Elasticsearch or OpenSearch
- Where
- Menu bar icon › Extensions… › Elasticsearch.
- You supply
- A name (for example logs), the URL such as
https://host:9200, the API key (or a user and password), and the CA certificate if the cluster uses your organization’s own. - Success
- Test Connection shows Connected., then the product and version, the cluster name, health, nodes and shards, and that the account cannot write or administer. Press Save Connection.
- Administrator
- Only for network access.
If it fails
The test says what to fix: a password, a host that cannot be reached, a certificate, or an account refused because it could write or administer. Common errors · Connection problems.
Kubernetes
- Where
- Menu bar icon › Extensions… › Kubernetes.
- You supply
- A name (for example prod), and a kubeconfig context or a service-account token with the cluster’s CA certificate. Turn on Read pod logs only if the role allows it.
- Success
- Test Connection shows Connected., then the platform, version, node count and account, and what the account reads: the whole cluster or some namespaces, and whether logs are allowed. Press Save Connection.
- Administrator
- Only for VPN access to a private cluster.
If it fails
The test says what to fix: a password, a host that cannot be reached, a certificate, or an account refused because it could write or administer. Common errors · Connection problems.
Prometheus or Thanos
- Where
- Menu bar icon › Extensions… › Prometheus.
- You supply
- A name (for example prod), the URL, a token if needed, and whether it is Prometheus or a Thanos Querier. Press Save Connection, then Test Connection. Build the map before asking: without an embedding service, enter the technologies you need (for example mysql, node) under Refresh technology and press Refresh. With an embedding service configured, Build Index maps the whole estate. Both paths.
- Success
- Test Connection shows Connected., then the backend, version and number of series. The refresh or index build completes for the jobs you will investigate.
- Administrator
- Only for network access.
If it fails
The test says what to fix: an unreachable endpoint, a certificate, rejected credentials or missing read access. Use the source’s least-privilege access instructions; this test does not certify every account capability. Common errors · Connection problems.
Splunk
- Where
- Menu bar icon › Extensions… › Splunk.
- You supply
- A name (for example prod), the management URL such as
https://host:8089, and a token (or a user and password). - Success
- Test Connection shows Connected., then the server name, Splunk version, the account’s roles and how many indexes it can read. Press Save Connection.
- Administrator
- Only for access to the management port.
If it fails
The test says what to fix: an unreachable endpoint, a certificate, rejected credentials or missing read access. Use the source’s least-privilege access instructions; this test does not certify every account capability. Common errors · Connection problems.
5. Attach it to a conversation
PostgreSQL
- Where
- Menu bar icon › Explain / Ask… with nothing selected. Type
/pg orders, using the name from step 4. - Success
- Verbiqo shows what it reached.
/sourceslists what is attached. - Administrator
- No.
If it fails
An unknown name lists the names you saved. A question reaches only what is attached in this window.
MySQL
- Where
- Menu bar icon › Explain / Ask… with nothing selected. Type
/mysql shop, using the name from step 4. - Success
- Verbiqo shows what it reached.
/sourceslists what is attached. - Administrator
- No.
If it fails
An unknown name lists the names you saved. A question reaches only what is attached in this window.
SQL Server
- Where
- Menu bar icon › Explain / Ask… with nothing selected. Type
/mssql sales, using the name from step 4. - Success
- Verbiqo shows what it reached.
/sourceslists what is attached. - Administrator
- No.
If it fails
An unknown name lists the names you saved. A question reaches only what is attached in this window.
Redis
- Where
- Menu bar icon › Explain / Ask… with nothing selected. Type
/redis cache, using the name from step 4. - Success
- Verbiqo shows what it reached.
/sourceslists what is attached. - Administrator
- No.
If it fails
An unknown name lists the names you saved. A question reaches only what is attached in this window.
Elasticsearch or OpenSearch
- Where
- Menu bar icon › Explain / Ask… with nothing selected. Type
/es logs, using the name from step 4. - Success
- Verbiqo shows what it reached.
/sourceslists what is attached. - Administrator
- No.
If it fails
An unknown name lists the names you saved. A question reaches only what is attached in this window.
Kubernetes
- Where
- Menu bar icon › Explain / Ask… with nothing selected. Type
/k8s prod, using the name from step 4. - Success
- Verbiqo shows what it reached.
/sourceslists what is attached. - Administrator
- No.
If it fails
An unknown name lists the names you saved. A question reaches only what is attached in this window.
Prometheus or Thanos
- Where
- Menu bar icon › Explain / Ask… with nothing selected. Type
/prom prod, using the name from step 4. - Success
- Verbiqo shows what it reached.
/sourceslists what is attached. - Administrator
- No.
If it fails
An unknown name lists the names you saved. A question reaches only what is attached in this window.
Splunk
- Where
- Menu bar icon › Explain / Ask… with nothing selected. Type
/splunk prod, using the name from step 4. - Success
- Verbiqo shows what it reached.
/sourceslists what is attached. - Administrator
- No.
If it fails
An unknown name lists the names you saved. A question reaches only what is attached in this window.
6. Ask a first question
PostgreSQL
- Ask
- The app is slow, what is the database doing? More: what to ask PostgreSQL.
- Expect
- Verbiqo samples active sessions for 10 seconds, then reports blocking with the chain of who waits on whom, and the share of time spent on lock and disk waits.
- Success
- Steps appear as Verbiqo works, then the answer, then its findings with the numbers behind them.
- Administrator
- No.
MySQL
- Ask
- The orders app is hanging, what is going on? More: what to ask MySQL.
- Expect
- Verbiqo samples sessions, lock waits and transactions for 10 seconds. A schema change queued behind an idle transaction is one finding, naming the session to end.
- Success
- Steps appear as Verbiqo works, then the answer, then its findings with the numbers behind them.
- Administrator
- No.
SQL Server
- Ask
- The sales app is hanging, what is going on? More: what to ask SQL Server.
- Expect
- Verbiqo samples requests and waits for 10 seconds. A blocking chain is one finding, naming the session at its head.
- Success
- Steps appear as Verbiqo works, then the answer, then its findings with the numbers behind them.
- Administrator
- No.
Redis
- Ask
- Redis is slow right now, why? More: what to ask Redis.
- Expect
- Verbiqo samples the counters for 10 seconds: commands a second, how busy the main thread is, evictions and refused connections. Then it reads the slow log, latency events and clients with large buffers.
- Success
- Steps appear as Verbiqo works, then the answer, then its findings with the numbers behind them.
- Administrator
- No.
Elasticsearch or OpenSearch
- Ask
- Review the cluster. More: what to ask Elasticsearch or OpenSearch.
- Expect
- The cluster review: unassigned shards with the cluster’s own reason, then disk, heap and rejections. Shard sizing, mappings, lifecycle and snapshots follow.
- Success
- Steps appear as Verbiqo works, then the answer, then its findings with the numbers behind them.
- Administrator
- No.
Kubernetes
- Ask
- Why is checkout crashing? More: what to ask Kubernetes.
- Expect
- Crash loops, out-of-memory restarts, image pulls, missing Secrets or ConfigMaps, and pending pods with the scheduler’s reason.
- Success
- Steps appear as Verbiqo works, then the answer, then its findings with the numbers behind them.
- Administrator
- No.
Prometheus or Thanos
- Ask
- Is anything down right now? More: what to ask Prometheus or Thanos.
- Expect
- What was checked, for which hosts and over which time range, the measurements the answer rests on, and what could not be checked.
- Success
- Steps appear as Verbiqo works, then the answer, then its findings with the numbers behind them.
- Administrator
- No.
Splunk
- Ask
- Any problems in index shop in the last 15 minutes? More: what to ask Splunk.
- Expect
- Known events per host, such as crashes, failovers and 5xx errors, with a timeline. The answer shows every search it ran and the exact period.
- Success
- Steps appear as Verbiqo works, then the answer, then its findings with the numbers behind them.
- Administrator
- No.
If it fails
A step marked refused says why. If the AI service times out, raise Timeout (ms). If it gives up early, check the model (step 2). How to ask.
7. Read the evidence
- Findings are Verbiqo’s, computed from what it read, each marked Critical, Warning or Note with the numbers behind it. The explanation above them is your model’s.
- Recommendations are proposals. Each has a risk and what to check first. Nothing is applied for you.
- Times. Counters cover the time since they were last reset, and the finding says since when. Questions about a time of day are read against the range the answer states: check it matches what you meant.
- Missing data is listed under what could not be seen. It is not the same as nothing wrong.
- No single cause is a valid result. A problem that has passed, or one the data does not cover, may end with what to look at next instead.
Reading the answer · Reading findings
8. Ask a follow-up
- Where
- The same window, for example what about locks? or continue.
- You supply
- The next question. It keeps the same system and time unless you name new ones.
- Success
- Continue resumes the latest investigation, preserving its scope and time range and reusing applicable evidence. It reads further when needed. Name a new time or scope explicitly when you want to change them.
- Administrator
- No.
If it fails
To start something unrelated, type /clear. To stop reading the system, type its command with off, such as /pg off.