Metrics and logs
Prometheus and Thanos · Splunk · Elasticsearch and OpenSearch
- Explain a latency or error spike from the measurements over time
- Find what a service logged when it started failing
- Review a search cluster: shards, disk, heap, lifecycle
Private AI for understanding systems · Beta
Understand what’s happening across your systems.
Ask about your applications, infrastructure, databases, metrics and logs. Verbiqo reads what you connect, read-only, and answers with the evidence it measured, using a model you choose.
It also rewrites, summarizes and explains text wherever you write. The model runs on your computer, behind your organization’s gateway, or as a cloud API you approve.
macOS 13 or later, Apple Silicon and Intel · Windows beta on request · 30-day trial, no account
- Databases
- Kubernetes
- Metrics
- Logs
Verbiqo reads only the sources attached to the question. This one used the Kubernetes cluster and its container logs.
-
Question
Why is checkout in the shop namespace failing?
-
Checks
- Found shop/checkout in the cluster’s map
- Read the namespace’s workloads and events
- Read the last lines of 2 failing containers, masked
- Checked the rest of the cluster for node trouble
-
Evidence measured by Verbiqo
CrashLoopBackOff2 of 2 pods, 94 restarts, exit 1 at 04:04 UTCLog line
DB_HOST unreachable:db01does not resolveCredentials in a logMasked before the model saw them -
Explanation written by your model
The checkout deployment cannot reach its database: the host name it connects to,
db01, does not resolve, so each container exits and restarts.
What an answer looks like
Evidence first, explanation second.
Every answer separates what Verbiqo measured from what your model wrote about it, and says what it could not see. Pick an example.
Kubernetes, with container logs · real transcript, lab cluster
Question
Why is checkout in the shop namespace failing?
Checks Verbiqo ran
describe deployment shop/checkout- Logs of the previous container, masked as read
- Warnings and node conditions across the cluster
Findings Measured
- Critical: container
appkeeps restarting (CrashLoopBackOff), 2 pods of 2, 94 restarts. - Critical: a container writes credentials into its log. Verbiqo masked them.
Explanation Your model
The application cannot reach the database: the log shows DB_HOST unreachable, and db01 does not resolve.
Limits stated
4 other findings on the cluster were kept apart from the question. Nothing was changed.
SQL Server 2022 · edited excerpt from a real transcript, lab server
Question
Orders are timing out in the shop database, what’s going on?
Checks Verbiqo ran
- Requests and blocking, sampled once a second for 10 seconds
- The heaviest statements and the indexes the optimizer asked for
- Deadlocks recorded by system_health, statements masked
Findings Measured
- Warning: the transaction log of shop is 65% full, with no log backup in the last day.
- Note: 2 deadlocks on
dbo.accountsin the last week.
Explanation Your model
Two deadlocks were recorded on dbo.accounts in the last week; these can abort transactions. The sample captured no active or blocked requests, so it does not establish the cause of the reported timeouts. The log is 65% full and its backup history needs review.
Limits stated
No requests were active or blocked during the sample, so a timeout happening at another moment would not show in it.
Prometheus · illustrative sample in Verbiqo’s output format, not a customer system
Question
Why did the orders database slow down between 14:00 and 17:00?
Checks Verbiqo ran
- Found the hosts and jobs the question means in its map of your Prometheus
- Measured connections, CPU and latency per host over the window
- Named hosts that were not monitored or had no data
Findings Measured
- Checked 4 hosts, 14:00–17:00.
- db02 reached 200 of 200 connections at 15:10.
Explanation Your model
db02 reached its connection limit at 15:10. Connection exhaustion may explain the slowdown, but the figures shown do not establish that requests queued or how the other hosts behaved. Connection-wait and latency measurements would help test that explanation.
Limits stated
Only what your Prometheus already collects can be measured. Verbiqo does not collect or store metrics itself.
What you can investigate
Three areas, one assistant.
Verbiqo reads the monitoring, logging and systems you already run. It sends no telemetry, raises no alerts and changes nothing. Conversations are not kept after you close them. On your computer it keeps a record of each read, for audit, and recent readings to compare a follow-up against.
Infrastructure and Kubernetes
Kubernetes clusters · machines through Prometheus
- Understand why a deployment or pod is failing
- Find the rollout that broke a service, and the command to go back
- Check room and risk before a drain or an upgrade
Databases
PostgreSQL · MySQL · SQL Server · Redis
- See what is slow or blocked right now, and why
- Check replication and availability before it becomes an outage
- Tune to the machine, then check whether the change helped
How it works
Connect, ask, inspect.
-
1
Connect an authorized source
Add a read-only account for each system, with its credentials kept in the macOS Keychain or Windows Credential Manager. Database and cluster connectors check for write or administrative privileges. For Splunk, use a least-privilege search role; Verbiqo restricts the searches it sends. Attach the sources a conversation may read.
-
2
Ask in plain words
Verbiqo works out which checks the question needs, runs them itself, and lets the model read further through the same rules. A name that matches several things is asked back, not guessed.
-
3
Inspect findings and evidence
Under every answer: the findings Verbiqo measured, which hosts and times were covered, and what could not be seen. The explanation on top is your model’s. Verbiqo checks it against the evidence where it can, and keeps its own findings beside it.
Who it helps
Tasks, not titles.
Who benefits depends on the access your organization grants and the sources it connects.
Find why a deployment fails
Developers and platform teams, with read access to the cluster.
Explain a latency spike
SRE and operations, from the metrics Prometheus already collects.
Answer “is it us?” with evidence
Support engineers, from the logs a service wrote.
Check database health before a release
QA, DBAs and release managers.
Review security settings for an audit
Security teams, with checks mapped to CIS Controls, ISO 27001 and SOC 2.
Summarize an incident for others
Anyone writing it up, from the thread or the notes.
Writing and summarizing
The same assistant, wherever you write.
Select text in any application and press one shortcut. Rewrite it clearer, shorter or more formal, summarize it or pull out the key points, explain an error, draft a reply, translate between English and Spanish, or sanitize a log before you share it. Summarize a whole document from a file, or open Ask and think a problem through.
- Incident write-upsKey points from a long thread: decisions, owners and dates.
- Support repliesA draft that matches the tone of the message, without inventing commitments.
- Share safelySanitize replaces hosts, addresses, keys and credentials with placeholders, by rule, with no model involved.
- Only what you approved changesEmails, links, paths, hosts and versions are checked after a rewrite. If the model altered one, the replacement is refused.
When an application only accepts copy and paste, Verbiqo warns first and lets you turn the fallback off. Clipboard-history tools may retain the original selection. Read the clipboard disclosure.
Writing and summarizingHi Alex,
Just checking if you saw my last message and can get back to me soon.
Thanks.
Verbiqo
⌘⇧R
Professional rewrite
I’m following up on my previous message. Please let me know when you have an update.
Your model
Replaces only the selected text
Security, privacy and deployment
Read-only, recorded, and on a model you choose.
What Verbiqo reads, where it goes, and what is kept.
Your systems
Read through a read-only account. Every statement, command or request is checked before it is sent. Secret stores are never read, such as Kubernetes Secrets or a server’s password settings. Credentials that turn up in data Verbiqo may read, such as a token in a log line, are masked before the model sees them.
Verbiqo, on your computer
Runs the checks, computes the findings, and records every read in a tamper-evident log that can be forwarded to your SIEM.
The model you chose
Receives your question and the masked evidence. With a model on your computer (Ollama), this stays on your computer. Through your organization’s gateway or a cloud API, it goes to that endpoint. Cloud providers need an explicit acknowledgement first.
What reaches Verbiqo’s servers
Licence activation and trial checks (the licence, a one-way device ID, the computer’s name for the seat list, the version) and two published files it reads for updates. Never your questions, your text or what was read. Privacy policy.
What is kept
Conversations end when you close the window. The record of reads, and readings kept to check whether a change helped, stay on your computer.
What it can change
Nothing in your systems. Verbiqo recommends a statement or command, with its risk and what to check first. A person decides whether to apply it.
Limits per question
Organizations can cap queries, seconds and questions per hour for each system, so many users never become a load on production.
For organizations
A signed policy decides which systems, features and models are allowed, and what may leave. It cannot be edited on the laptop. Licences deploy with Jamf, Intune or Group Policy, online or fully offline. Security reviews map Verbiqo’s checks to CIS Controls v8, ISO 27001 and SOC 2: evidence for an audit, not a certification.
Supported sources and setup
What it connects to, and what it needs.
Sources
- Metrics
- Prometheus, Thanos
- Logs and search
- Splunk, Elasticsearch, OpenSearch
- Infrastructure
- Kubernetes
- Databases
- PostgreSQL, MySQL 8.0 and later, SQL Server 2016 to 2022, Redis and Valkey
Applications are read through their logs, their metrics and the clusters they run on. Verbiqo does not connect to application code or APM tools.
Models
- Local
- Ollama
- Organization
- OpenAI-compatible gateways such as LiteLLM or vLLM
- Cloud
- OpenAI-compatible APIs and the Anthropic API, after an explicit acknowledgement
Platforms
- macOS
- 13 or later, Apple Silicon and Intel: signed and notarized
- Windows
- 10 and 11 (x64): unsigned beta, given out on request. Not downloadable here yet
- Linux
- Command line tool, x64 and arm64, on request with a team or organization licence
You need
- A model endpoint: local, your organization’s, or an approved cloud API
- A read-only account per system. The documentation gives each one
Try it or pilot it
Start on one Mac, or with your team.
One notarized app: a 30-day trial that becomes licensed when you import the signed licence you receive. Prices are agreed with you. None are published.
30-day trial
On this Mac
- Every writing action and document summaries
- Every supported source, read-only
- Security review and the record of reads
- Ollama, OpenAI-compatible endpoints or the Anthropic API
Guided pilot
For a team
- Set-up with your sources and model
- Signed policy, limits and the record of reads
- Licences by Jamf, Intune or Group Policy, online or offline
- Deployment and security review
Personal licence
Keep using the same app
- No separate licensed download
- No Verbiqo account
- Activates online once, then checks in quietly
- Bring your own model or API account
Verbiqo v0.1.25 beta
Try Verbiqo on your Mac.
Signed with a Developer ID and notarized by Apple. After 30 days, import a signed licence into the same app to continue.
Verbiqo.appDeveloper ID signed · Apple notarized
Pilots, licences and access
Tell us what you need.
We reply by email to agree the sources, the model and the scope of a pilot on your own systems.
Questions
Straight answers.
What it does, how your data is handled, and what to expect.
What can I do with Verbiqo?
Ask questions about the systems you connect, and get answers built on what Verbiqo reads from them. It covers metrics and logs (Prometheus, Splunk, Elasticsearch and OpenSearch), Kubernetes clusters, and PostgreSQL, MySQL, SQL Server and Redis. It also rewrites, summarizes and explains text in the apps you already use. Start with the documentation.
How does it investigate applications?
Through the logs, metrics and infrastructure your applications run on. Verbiqo reads what a service wrote to Splunk, Elasticsearch or OpenSearch, the measurements Prometheus holds for it, and the state of its pods in Kubernetes. It does not read source code or connect to application performance monitoring (APM) tools. How to ask about your systems.
Does it replace my monitoring or logging tools?
No. It reads the monitoring, logging and systems you already run, when you ask a question. It does not collect telemetry, watch continuously or send alerts. On your computer it keeps only a map of what your Prometheus collects and readings used to check whether a change helped. How Verbiqo reads your systems.
Can it change my connected systems?
No. Investigation requests are checked before they are sent and restricted to reads. Database and cluster connectors also check for write or administrative privileges. Splunk searches are restricted by Verbiqo; configure a least-privilege search role because its connection test does not certify every capability of that role. The model can only ask Verbiqo for more reads. It has no shell or file access on that path. Writing actions replace only the text you selected, after checking the result, and a document summary reads only the file you choose. What read-only means here.
Where does my data go, and what is stored?
Your question and what Verbiqo reads go to the AI model you configured, and to no other service. With a model on your computer, such as Ollama, that stays on your computer. Verbiqo still connects to the sources you attach, and to verbiqo.app for update checks and licence or trial checks, which carry no content. Conversations are kept in memory until you close them. A record of every read stays on your computer for audit. Passwords go in the macOS Keychain or Windows Credential Manager. What goes where, and how long it is kept.
Can its answers be wrong?
Yes, the explanation can be. Verbiqo computes its findings from what it read and shows them beside the answer, so you can check the explanation against them. The explanation is written by your model, which can misread evidence. Verbiqo checks figures and some claims against what it measured where it can, and says when evidence is missing. An investigation that ends without a single cause can still be a correct answer. How to read an answer.
How is it different from an AI assistant using MCP servers?
MCP is a protocol for connecting AI applications to tools and data. Verbiqo is a product that arrives with the checks for each system, read-only enforcement, masking and a record of every read already in place, so a first investigation needs a read-only account rather than an integration project. How much an MCP-based set-up includes depends on the product or servers you choose. Verbiqo as an MCP server is planned, not available. The fuller comparison.
What do I need to get started?
A Mac with macOS 13 or later, and an AI model: on your computer, your organization’s gateway, or a cloud API you approve. Investigating systems needs a capable model with a large context window, and a read-only account for each system, which an administrator may need to create. Writing needs only the model and the macOS Accessibility permission. First investigation · First writing task.
Which platforms, models and licences are available?
macOS 13 or later on Apple Silicon and Intel is signed, notarized and downloadable. Windows 10 and 11 is an unsigned beta given out on request, and the Linux command line is available on request with a team or organization licence. Models: Ollama, OpenAI-compatible endpoints and the Anthropic API. After a 30-day trial you import a signed licence. Most activate online once and then check in quietly, at least once every 37 days, and offline licences exist for computers without internet access. Licences · Tested models.