Troubleshooting
Find the problem you are seeing, then follow the fix. If it is not here, see Still stuck.
The shortcut does nothing
Check Verbiqo is running
Its icon should be in the menu bar (on Windows, by the clock). If not, open it from Applications, or from the Start menu.
Check Accessibility
In System Settings, then Privacy & Security, then Accessibility, switch Verbiqo off and on again.
Check for a clash
Another app may use the same shortcut. Choose a different one under Global shortcut in Settings.
“Could not reach” the AI service, or no models listed
- In Settings, press Test & Load Models and read the message it shows.
- Ollama: make sure Ollama is running and has a model: run
ollama pull gemma3, then test again. - Cloud or organization service: check the Endpoint and API key. A VPN or firewall may block the address.
- If the service reached but lists no models, type the model name yourself.
“macOS (or Windows) trusts this endpoint’s certificate, but Verbiqo’s engine does not”
Your AI service uses a certificate from your organization’s own authority. The system trusts it; Verbiqo’s engine reads only the certificate authorities it is given.
Get the CA file
Your organization’s certificate authority, as PEM text: the file starts with
-----BEGIN CERTIFICATE-----, whether it is named.crtor.pem. Your IT team has it. A binary (DER) file must be converted first:openssl x509 -inform der -in ca.crt -out ca.pem.Set it in Verbiqo
In Settings, Trusted CA, just below Endpoint: press Choose… (Browse… on Windows) and pick the file. Verbiqo uses it where it is and does not copy it, so keep it there.
Test again
Press Test & Load Models, then Save. The same file is then trusted for the AI service and every connected system.
In an organization, your administrator can set it for every computer (Trusted CA). A terminal on the same computer trusts it too, along with everything in the computer’s trust store. Without the app, set NODE_EXTRA_CA_CERTS to the file.
My text was not replaced
- You moved the cursor or edited the text while Verbiqo worked. It stops rather than write in the wrong place. Select again and retry.
- The model changed something protected, such as a link, address or version number, so Verbiqo refused the result. Try again or pick another action.
- The app does not allow direct changes and the copy-and-paste fallback is turned off. Turn on Allow keyboard/clipboard fallback for compatible apps in Settings.
A document cannot be read
| Message | Fix |
|---|---|
| …does not contain readable text | It is probably a scanned PDF. Run it through text recognition (OCR) first. |
| Unlock the PDF… | Remove the PDF’s password, then try again. |
| …larger than the 50 MB safety limit | Use a smaller file, or split it. |
| …more than 150,000 characters | Summarize one part at a time. |
A connected system will not connect
- Press Test Connection in Extensions and read the message.
- Splunk: use the management port, usually
8089, not the web address on port 8000. Can read no indexes means the account needs permission to search. - An account refused as able to write or administer: create the monitoring account from the set-up page for PostgreSQL, MySQL, SQL Server, Redis, Elasticsearch and OpenSearch or Kubernetes.
- Certificate errors: see below.
- Prometheus: if questions miss obvious services, press Build Index again, or Refresh the services by name.
- No Extensions… in the menu? Turn on Let Verbiqo read databases and other systems in Settings. If the setting is not there or cannot be changed, your administrator has turned it off.
Certificate and TLS errors
The test says what to change under the error. The usual cases:
| Message | Fix |
|---|---|
| self-signed certificate, unable to verify, not signed by a CA this computer trusts | Choose the CA certificate in the connection, or name it once under Settings › Trusted CA (below Endpoint). For a self-signed lab server, TLS mode Require (or tick Insecure). |
| does not match certificate’s altnames | Connect with the host name on the certificate, or have the certificate reissued for the correct name. Keep Verify full for production; Verify CA omits host-name verification. |
| no encryption, requires secure transport | The server accepts only TLS: choose Require or stricter. |
| does not offer TLS | Enable TLS on the server. In a lab only, choose Prefer (PostgreSQL, MySQL) or Optional (SQL Server), which connect without encryption. Redis has no Prefer: for a port that does not serve TLS, choose Disable, or connect to its TLS port. |
| requires a valid client certificate | Add the client certificate and key to the connection. For PostgreSQL and MySQL a key protected by a passphrase works: enter it in Key passphrase, or in a terminal --ssl-key-passphrase-env VAR or --ssl-key-passphrase-stdin. For other systems the key must not have one. |
| certificate has expired | Renew the certificate on the server. |
An answer about my systems looks wrong
- Check the hosts and the times listed in the answer. If they are not what you meant, ask again naming the service and the time exactly.
- If Verbiqo picked the wrong thing, reply to its Did you mean with the right one.
- Read the coverage note: hosts that were down or had no data are not part of the result.
Questions about my systems: common messages
| You see | Why | Fix |
|---|---|---|
| The investigation stopped early, or a question gives up before finishing | Timeout (ms) also limits a whole Prometheus investigation, and the default is 30 seconds. | Set Timeout (ms) to 300000. Behind a gateway, raise its timeout too. More |
| Evidence was reduced to fit a …-token budget with a small number | The model server did not report its context window, so Verbiqo assumed 4,096 tokens. | Set Context window to the server’s real size, such as 32768. More |
| Every step takes minutes; the model thinks before it answers | A reasoning model, or a hybrid one with reasoning on. | Use an instruct model. Reasoning models |
| The same question gives a different explanation each time | No temperature set, so the server’s default (often 0.7) applies. | Set Temperature to 0.2. More |
| … matched no job in Verbiqo’s map of this server | The name you used is not a job Verbiqo knows, so it carried on with the previous subject. | Name the job as Prometheus labels it, or Refresh that technology in Extensions so the map has it. |
| That selects … series, above the ceiling of 1,000 | A metric with a series per table, index or user is too wide to read in one request. Verbiqo tries one host at a time first. | Nothing to fix: it protects Prometheus. The answer says that measurement was not made. |
| The command line or model check says the model is unavailable | The model server uses your organization’s certificate, which is neither in this computer’s trust store nor in Settings › Trusted CA. | Add it to either, then retry. For one shell: export NODE_EXTRA_CA_CERTS=/path/to/your-ca.crt (Windows: set …). |
| The local Prometheus index ran out of room, or reached its … MB limit | A large server’s index needs more disk than is free (it keeps 2 GB free), or more than your administrator’s limit. | Free some disk, or raise VERBIQO_PROMETHEUS_INDEX_MAX_MB, then build again: it continues where it stopped. Large estates |
| Embedding endpoint returned HTTP 400 … A text was longer than the embedding model accepts | The embedding model’s input limit is shorter than a job’s description. | On vLLM, start it with --max-model-len 4096 or more. Example |
| Embedding endpoint returned HTTP 401 … refused the key | The embedding service’s key expired, or is not allowed the embedding model. | Save a new key under Extensions › Prometheus › Semantic index, or turn on Use the model’s API key if your gateway gives that key access to the embedding model too. |
| Splunk: lines … stamped about N hours … | Splunk reads that log’s timestamps in the wrong time zone, so its times are off by those hours, in Splunk’s own search too. | Ask your Splunk administrator to set TZ for that sourcetype in props.conf. |
| Splunk: … assumed, the question named no period | The question had no period Verbiqo understands, so the last 24 hours were searched. | Add one: in the last 2 hours, since 9am, yesterday. A time needs am, pm or minutes. Periods |
| Splunk: refused the session key … | Your Splunk address fronts several search heads, and a session key works only on the one that issued it. | Use a Splunk token for the connection, or a directory login with its password saved; in a terminal chat, type the password when asked. |
| The model endpoint refused the API key (HTTP 401) | The key has expired or been revoked. Keys from a sign-in portal such as Keycloak last a set time. | Get a new key from whoever issued it and save it in Settings › API key. The terminal uses the same saved key, or set VERBIQO_API_KEY for one shell. |
Windows: an upgrade asks me to close Verbiqo
An upgrade to 0.1.21 or later (installer or choco upgrade verbiqo) closes a running Verbiqo itself, whatever version it is, and starts it again when done. If an upgrade still stops on files in use, quit Verbiqo from its icon by the clock (Quit Verbiqo) and run the upgrade again.
Licence problems
- The licence is already active on all its devices: deactivate one of the computers Verbiqo lists, or on the old computer press Deactivate in Settings.
- Could not reach verbiqo.app to activate: check your connection and try again. Once activated, Verbiqo works offline for up to 37 days.
- Trial ended: import a licence in Settings › Licence to continue. Licences
- Settings › Licence says Managed by your administrator and Import is off: your organization deploys the licence. To move to another computer or free a seat, ask your administrator.
- Not renewed: all seats in use: another computer took this one’s seat while it was away. Press Free a Seat…, or ask whoever manages the licence.
Still stuck
Turn on the diagnostic log
In Settings, turn on Write a diagnostic log. It never records your text.
Repeat the problem
Do what went wrong once more.
Send it to us
Press Show Log and email the file to hello@verbiqo.app, with what you did and what you saw. Do not include passwords or confidential text.