Reference

Troubleshooting

The handful of things that come up, and how to resolve them. If none of these fit, the red support button under Help in the assistant reaches a person.

A licence warning is showing

The assistant stays licensed by checking in with Naxis. A warning means it hasn’t been able to reach us recently — almost always the server has lost outbound internet.

• Confirm the server can reach the internet (a domain doesn’t matter here — this is outbound, not inbound).
• Check any firewall or proxy allows HTTPS out to naxistechnologies.com.
• Once connectivity is back, the warning clears on the next check-in. You have a long grace period, so a short outage is harmless.

People can’t reach the assistant / no HTTPS

This is about the public address, which is separate from licensing.

• Self-hosted: point a domain’s A record at the server and re-run the installer — HTTPS is issued automatically. Then set that domain as your public address in Settings. See Install on your server.
• Managed: your address is provided by Naxis with HTTPS already in place. To use your own domain, ask your Naxis contact.

Invitation or reset emails aren’t arriving

• Check Settings → Email (SMTP) is filled in, and use Send test.
• Confirm the mailbox credentials and that the provider allows SMTP sending.
• You can always hand someone their activation link directly — it’s shown in the console when you invite them, so a mail problem never blocks onboarding.

Full setup: Email (SMTP).

Someone can’t sign in

• Forgot their password: self-serve from the sign-in page — a single-use link, valid 24 hours, emailed to the account. No email configured yet? Copy the reset link for them from the Users page instead.
• The administrator sign-in code (2FA) never arrives: this isn’t a lockout — if the mail server can’t be reached, sign-in falls back to password only automatically, and the fallback is recorded in the audit log. Fix Settings → Email (SMTP) when convenient, not urgently.
• Invite expired before the person used it: reissue it from the Users page — the old link simply stops working, nothing else is affected.

New documents aren’t showing up in answers

• Indexing runs on a schedule, and after every manual Sync. Check Background activity — a sweep may still be in progress.
• Open the source under Documents and find the file. Status should read active; failed shows a red reason why, and a Passages count of 0 usually means the file came through empty or unreadable. Anything failing is also collected under Needs attention.
• Confirm the document is inside a connected source and in a supported format.
• Confirm the asker has access to that source — if they don’t, they won’t see it in answers even though it’s indexed.

Full pipeline: How documents become searchable.

A source’s Test button fails

Test reaches the source with its current credentials and reports exactly what went wrong.

A chat channel is connected but never replies

Test checks your credentials against the platform — it doesn’t always confirm the platform is delivering messages here. Most channels (WhatsApp, Slack, a phone number, Teams and more) receive over a webhook: an address shown behind the channel’s Setup button once it’s created.

• Paste that exact address into the platform’s own configuration (Slack’s Event Subscriptions, Meta’s WhatsApp webhook form, Twilio’s “a message comes in”, and so on) — passing Test doesn’t do this step for you.
• The address is built from your public base URL (Settings). Change that URL later and the platform side needs re-pasting too.
• Viber is the exception: it registers its own webhook when you press Test, so a failure there names the real problem — almost always the public base URL.
• No public address available? Telegram picks up its messages on its own, and Slack’s Socket Mode connects out instead of waiting for a webhook — see each channel’s setup guide (the ? next to its description in Chat channels).

An answer was refused, but the information exists

This is the assistant working as intended, not a fault: it answers only from material the asker can read, and would rather say so than guess. A refusal always means one of two things — the document isn’t in yet (not connected, or not synced), or this particular person doesn’t have access to it.

• Check the document reads active under Documents for the right source (see above).
• Open Users, find the person, and check their access — “Check a person’s view” (also shortcut-linked from Overview) shows exactly what they can and can’t read, computed the same way the chat itself decides.
• Fix whichever it is — connect or sync the source, or add the person’s group — and ask again.

More on why: How answers are made (and why they can be trusted).

An update seems stuck

Updates back up first, health-check, and roll back automatically on failure — see what happens during an update. That makes a genuinely stuck update rare. Before contacting support:

• Deployments update themselves — managed and self-installed alike — by default on the coming Sunday at 02:00 in the deployment’s own local time, or sooner from Settings → Software updates (Update now). If an error page appears during that window, wait ten minutes and reload — that’s normal downtime, not a stuck state. See how updates arrive.
• A deployment run outside the update channel, with no licence link of its own, shows no Software updates card at all — it follows whatever process you put around it, so there is nothing here to look stuck. A server you installed from your own Naxis link is not that case: it updates on the schedule above.
• Still down well past that window? Don’t intervene on the server yourself — contact support and they’ll look; the outcome is already visible to Naxis.

Last updated 3 Sep 2026