Standby

Whole-server backups, and a weekly drill that restores each server on the backup server and boots it.

Added

• Whole server backup preset. Everything on the disk except /proc, /sys, /run, scratch space and package caches, plus database dumps: what a server is rebuilt from.
• Weekly recovery drill that boots each server. The backup server restores each server's newest whole-server backup, checks it is a working system, boots it with no network, measures the restore and projects the time to be back on a fresh server, then deletes the copy.
• Can each server come back? on the Backups page. Last backup, last drill, its verdict in words, the projected time back on a fresh server, and Drill now.

Upgrading

Switch backup policies to Whole server once the backup server has room (Move to volume and Grow from 0.16.69). Allow scripts for the backup server on Host access: the drill runs there.

The backup server keeps backups on its data volume, not its OS disk.

Added

• Move to volume and Grow on the Backups page. When a backup server keeps its repository on the OS disk beside an unused volume, one button moves it, checking every file arrived before removing the old copy. When the volume was made bigger, Grow extends the filesystem online.

Fixed

• Backups were kept on the backup server's OS disk. The installer now mounts an attached, unused volume for the backups (formatting it only when it is blank) and refuses the OS disk unless ALLOW_ROOT_FS=1.
• Backup capacity can follow a volume. The Backups page no longer rewrites the repository's path to the disk that held it, which pinned capacity to the OS disk.

Upgrading

Allow scripts for the backup server on Host access, then press Move to volume on the Backups page.

The "Auth & intrusion events" score now counts problems you can fix, not the amount of attack traffic.

Changed

• The tile lists what lowers the score. Each condition is named with the servers it applies to and what it costs.

Fixed

• The intrusion score no longer falls to 0 on a busy week. It was scored from the number of open high-severity events, so ordinary scanner traffic against a web firewall that only detects pushed it to 0. It now counts conditions: servers where attacks got past the web firewall (−15 each), attacker logins not yet triaged (−60 each), servers with a possible intrusion not yet triaged (−30 each), and other kinds of open high-severity events (−5 each, at most −30). Blocked attacks and the number of events cost nothing.

The incident page is rebuilt around one question: why did this fire?

Changed

• The incident page opens on why it fired. One sentence saying what happened, the facts behind it (what, where, measured, normal, limit), the evidence itself (the metric chart, the detected file and signature, the raw security event), what it usually means, and what Standby does not know. Then what happened since, the runbook, and how often it happened before. No tabs.
• How an incident ended is stated plainly. Resolved by a person with their note, recovered, marked not a problem, or closed because the source stopped reporting it, which is not the same as fixed.
• Postmortem, status page, war room, blast radius, mitigation, regulator reporting and AI help are under More. They no longer fill the page as separate cards.

Fixed

• Alert details sent to Standby are kept. The generic endpoint keeps annotations, description, message and details, and kstack events keep their message and detail, so the incident page can show them.

Runbooks are now suggested on the kinds of incident Standby actually raises.

Changed

• An incident with no runbook says so. It links to a new route already filled in with that incident's labels. Runbook routes is easier to read and is paged.

Fixed

• Incidents get a suggested runbook. Routes now match the incident itself: its alert's labels, its source, category and severity, and for security events (malware, reverse shells, logins) their type. New starter routes cover the incidents Standby raises most often. A route value of * means any value.
• The same route is no longer listed dozens of times. It was added again on every start. Exact copies are removed; the oldest is kept.

The top-of-page bar about SMS and voice not being set up is gone.

Changed

• No more bar about SMS and voice not being set up. Nothing was wrong: pages on those channels go by email and push instead, and Settings → Telephony still says so. The bar at the top of the page now appears only when paging is actually broken, or when a schedule has nobody on call.

Fleet updates now shows which updates install automatically and when, and maintenance windows are kept inside your maintenance hours.

Added

• Fleet updates says what happens on its own. The page opens with when windows run, the security deadline, how much notice customers get, and the policy each notice links to.

Fixed

• Updates that install automatically no longer ask to be approved. With the maintenance program on, each server's maintenance window installs every pending update. Updates found after a window was booked are now tied to that window and listed under Installs automatically, with the date and whether customers have been told.
• Maintenance windows stay inside your maintenance hours. A window is moved into your maintenance hours before customers are told about it. Windows that end up at the same time become one window, so subscribers get one email.

Upgrading

Check that the maintenance hours (Fleet updates → Auto-update) and the policy link (Maintenance windows) match what you promise customers. Windows not yet announced are moved into those hours at the next hourly pass.

Host access has a page of its own, with search, filters and pages.

Changed

• Host access has its own page under Fleet & endpoints. It lists which servers accept scripts and the web terminal, with a search, a filter and pages of 25. Each server has an on/off switch for scripts and one for the terminal. It used to be a small scrolling panel inside the Fleet command center.
• The web terminal says where to allow a server. When a server can't take a session because it isn't allowed, the New session dialog links to Host access.

Status-page subscribers now get an email before scheduled maintenance.

Fixed

• Status-page subscribers are emailed about scheduled maintenance. Maintenance windows were left off the status page when their servers were not linked to a status-page component. Most servers never are, so subscribers heard nothing. A window is now announced unless one of its servers is set to Hidden. The notice does not name any servers.
• One notice per maintenance time. When several servers are booked for the same time, they now share one window, so subscribers get one email instead of several.

Upgrading

Nothing to do. Within an hour of the update, Standby announces every scheduled window that was never announced, and merges windows at the same time into one. A window closer than the policy's notice period is moved to the earliest time the notice allows before it is announced. To keep a server's maintenance off the status page, set its status-page audience to Hidden.

Windows endpoints now report their installed software, listening ports, hardware and hardening settings. Windows event log errors and warnings keep their level.

Added

• Installed software on Windows. The agent reports what Windows lists under Settings → Apps, every 25 minutes. The Applications tab shows it with a filter, and installs, removals and upgrades appear on the Changes tab. Vulnerability matching for Windows applications is not available yet.
• Listening ports on Windows, with the program that owns each one. They appear on the Changes tab and are used by the firewall investigator.
• Hardware on Windows. Make, model, board, BIOS, processor, memory, disks and graphics appear in the asset register.
• A Windows hardening card on the Security tab. It shows Secure Boot, User Account Control, SMB1, the members of the local Administrators group, and whether Windows installs updates by itself, with the reason when it does not. Automatic updates and installed browsers now count for A.8.8.AU and Cyber Essentials.

Fixed

• Windows event log errors and warnings keep their level. They were all stored as information. This takes effect once the agent has updated its log shipping.
• A third-party firewall that is on is no longer graded "inactive". When a suite such as Norton turns Windows Firewall off and protects the machine itself, the firewall evidence names that product. A registration whose program is no longer installed does not count.

Upgrading

Update your Windows agents (Fleet → Update all agents) to agent 0.159.0.

Why

Several host pages were empty for Windows machines because the agent had no Windows source for them, and every Windows error in the logs was being filed as information.

Windows endpoints now report what is true. A Windows PC is only shown as fully patched after a confirmed online update check. Remote Desktop access closes on time. Several Windows readings that showed green without evidence now say "unknown" or show the real value.

Added

• "Refresh now" works on Windows. It runs the online update check immediately. Nothing is downloaded or installed.
• Windows hosts report time sync, the full Windows build and the firewall settings Windows actually enforces. Time sync counts toward A.8.17. The build includes its update number (for example 26200.9550). Firewall settings include those set by Group Policy.

Changed

• Windows pages say what Windows does not report yet. Load average is no longer shown as 0.00. The Applications and Changes tabs say plainly what is not reported on Windows yet. Disk encryption and backup evidence name BitLocker and Standby backups.

Fixed

• "Fully patched" no longer appears on an unchecked Windows update list. The agent asks Windows Update online once a day for the pending list, and asks Windows to rescan so its Settings page catches up. "Last Windows Update check" is now the time of that check. A list not confirmed in the last 30 hours shows "not confirmed patched". Patch management (A.8.8) uses the same rule on every host.
• Windows Settings no longer keeps showing updates the agent installed. After installing updates, the agent tells Windows to rescan.
• The restart warning clears once the machine has restarted.
• A failing agent update on Windows is rolled back, and a good one is confirmed. Before, the process that watches a new version could be stopped along with the old agent.
• Antivirus is no longer marked "not running" after a month without a restart. A product counts as installed when its own program is on disk. When the antivirus check cannot run, the host shows "unknown" instead of "no antivirus".
• A restart check that did not run is "unknown", not "no pending reboot".

Security

• Remote Desktop access closes on time. The agent enforces the deadline itself, so access granted before the machine slept or was switched off no longer stays open. A grant with no deadline on record is closed. The host page shows when Remote Desktop is on, the address it is open to and until when.

Upgrading

Update your Windows agents (Fleet → Update all agents) to agent 0.158.0. The first check after the update closes any Remote Desktop access an older agent left open with no deadline. To keep using Remote Desktop, open it again from the host page; the new access has a deadline.

Why

A Windows PC could show "Fully patched" while Windows had not checked for updates in weeks and security updates were waiting. Remote Desktop access could also stay open long past its deadline without anything saying so.

Updates no longer take Standby offline. The new release is prepared and switched in while the current one keeps answering.

Changed

• No "Updating" page during an update. Standby now runs two copies of its web service side by side during an update. The new release is prepared and its database changes are applied while the current release keeps answering. Once the new release passes its health check, new requests go to it, and the old release finishes what it already has and stops after a minute. System → Updates shows "no downtime · switch N s" for such a run.
• A failed update goes back without an outage. If the new release fails after the switch, requests go straight back to the old release, which is still running.
• Releases are checked so the switch is safe. A release whose database changes the previous release could not work with is refused before it ships, unless it updates behind the maintenance page instead. Before each release, the previous release's own code is also run against the new database layout.

Upgrading

Nothing to do. This update itself still uses the short "Updating" page, because the updater already on your server installs it; it prepares the server, and every update after it has no downtime. If the upstream standby_oncall block in your nginx site for Standby was edited by hand, updates keep using the short "Updating" page and the update log says why.

Why

An on-call and status platform has to stay reachable while it updates. Until now every update put up an "Updating" page; from the next update on, nothing is refused.

Standby's own file-integrity check no longer raises "Reverse shell" incidents on the hosts it scans.

Fixed

• False "Reverse shell" alerts from Standby's own file check. The host agent's file-integrity check lists the programs in /usr/bin. That list includes files named mkfifo, nc and bash, and the reverse-shell detector matched those names. It raised a high "Reverse shell" incident on the hosts the check ran on. A reverse shell is now reported only when those programs are run as commands, not when they appear as file names in a listing. Until a host's agent is updated, such reports still arrive, but they are kept on the threat feed at info level and no longer open an incident.

Notes you write on an incident are always kept and shown near the top of the page, and what you said about an address follows it to the other incidents about that address.

Changed

• Notes are at the top of the incident page. The note box used to be the last thing on the Overview tab, under the activity timeline. It is now a Notes card just under the incident header, showing what people wrote.
• Notes carry to incidents about the same address. The Notes card shows what was already said about the incident's IP address on other incidents, in "Not a problem" rules, and on the threat feed. It also lists the other incidents still open about that address and can resolve them with one note.
• "Not a problem" asks about the others. When other incidents about the same address are open, it asks whether to close them too, each with your note and marked "same as" this incident.

Fixed

• Notes written after Resolve were thrown away. Pressing "Not a problem" on an incident you had already resolved saved your explanation only on the security events behind it. The incident said "Resolved by …" and nothing else. The same happened to a resolve-with-note from the incident drawer, or a bulk "false positive", on an incident that was already closed. The note is now always written on the incident.
• The incident drawer hid people's notes. It showed the last five timeline rows, which were often "Additional signal attached" lines. Notes people wrote are now always shown, and a resolved incident still takes a new note.
• Threat-feed triage notes reach the incident. A note saved on a security event that paged also appears on that event's incident.

Updates take Standby offline for seconds instead of minutes, each update reports how long it was offline, and the bell keeps working afterwards.

Changed

• Shorter updates. Everything that can happen while the running version keeps serving now happens before the "Updating" page goes up: downloading, building, the safety backup of the database, certificate and permission housekeeping. Only the switch itself — stop, migrate, start — happens behind the page. The services stop in parallel with a 20-second limit, and live connections (the bell, chat, log tail) close at once instead of holding the stop open.
• Each update says how long Standby was offline. System → Updates shows "offline N s" for every run.

Fixed

• The bell keeps updating after an update. Its live connection now comes back by itself after the "Updating" page, instead of staying silent until the page was reloaded.

Why

An on-call console has to be there when you need it, including while it updates. This is the first step; a following release removes the offline moment altogether by starting the new version beside the old one.

A "Not a problem" view on the threat feed shows everything Standby learned from your notes, and lets you undo any of it.

Added

• See and undo what "Not a problem" taught Standby. The threat feed has a Not a problem view. It lists every rule Standby learned from your notes — what it is, the host, the address, what it matches, how often it has matched and when last — with Undo, and the SSH keys you confirmed the same way with Forget. After either, the next matching event is judged as if Standby had never been told. Undone rules stay listed under Show undone.

Why

Telling Standby that an alert is harmless should never be a one-way door. You can see each thing it learned, how often it has quietened an event since, and take it back.

Logins between your own servers stop paging, a new "Not a problem…" action teaches Standby what an alert really was, and acknowledging in a side panel shows on the list straight away.

Added

• "Not a problem…" teaches Standby. On the incident page and in the threat feed: say what it is, and Standby closes it, keeps your note and learns. A login key becomes known for that account on that server; anything else arrives quietly next time with your note. The rules it learns can be listed and undone.

Fixed

• Logins from your own servers no longer page. A service account on one of your servers logging in to another with its SSH key paged every few minutes as "a key Standby has never seen". A login from one of your own servers' addresses is now treated like an allow-listed address. A key added to authorized_keys behind Standby's back still pages from anywhere, and a key is no longer called "never seen" for an account whose keys Standby does not read.
• Acknowledge shows straight away. Acting in a side panel now refreshes the list behind it at once.
• During an update, actions say so. A click while Standby applies an update now says "Standby is applying an update — nothing was changed" instead of "Request failed".

Why

An alert that keeps firing about your own systems trains people to ignore alerts. This release stops the one that fired after 0.16.52 and gives you a way to explain any other so it does not come back.

Standby now sees what your servers talk to, what they run and which of their files change, watches your domains for hijacked nameservers, unexpected certificates and lookalikes, and spots pass-the-hash logons on Windows. Includes the host agent standby-agentd 0.157.0. Nothing in this release pages; every watch can be turned off per host.

Added

• See where each server sends traffic. The agent notes every outbound connection once a minute (program, user, address, port) and sends a summary each hour. The host's Security tab shows a Talks to list, and the pop-up for an address lists your servers that connected to it. A server calling an address your fleet has flagged as an attacker reaches the bell, and so does a lookup of a cryptocurrency mining pool. A connection on a mining-pool port, or a first connection to a country no server has talked to in 30 days, goes to the threat feed. Only destinations are kept, never contents.
• See what runs on each server. The agent checks new processes on the server itself and reports only matches: a shell started by the web server, a download piped into a shell or a program run from /tmp, a reverse shell, an unknown program using a CPU core for ten minutes, and Windows tools used to fetch or run code. A web-server shell reaches the bell when its command looks around, downloads or connects out. Each host has a Process watch card with an allow-list.
• Know which files change. The agent fingerprints each website's files, /etc and the system program folders. A new script file in a website reaches the bell, as do changes to accounts, sudo, SSH, PAM, cron or ld.so.preload, and a system program not owned by root or whose date was put back. Changes made just after Standby published a site file, installed a package or ran a script are marked as Standby's. Each host has a File integrity card.
• Watch your domains. A Domain watch card on the DNS page tells you when the nameservers of any zone your DNS providers serve change. For the domains you choose, it also checks Certificate Transparency daily for certificates from an issuer the domain has not used before, and looks for newly registered lookalike domains.
• Spot pass-the-hash on Windows. The agent reports NTLM network logons to administrator accounts and the logon type that runas /netonly and pass-the-hash tools create. The first such logon from an address in 30 days reaches the bell; a pair seen before stays quiet.

Fixed

• The account watch reads locked-down cron folders. Where the agent may not read /etc/cron.d, it now reads it through its root helper instead of reporting "partial".
• Cleaner ATT&CK coverage. The watches' first snapshots no longer count as unmatched attack events, and ClamAV detections count toward a technique.

Upgrading

• Update your agents (Fleet → Update all agents). Each watch starts with a quiet baseline.
• To watch certificates and lookalikes for a domain, add it on DNS → Domain watch. Nameservers are checked for every zone your DNS providers serve without any setup.

Why

Once an attacker is inside a server, the signs are a connection to somewhere new, a program the server never runs, or a file that appeared in a website. Until now Standby watched the front door. This release watches inside, while keeping alerts for the patterns that matter.

Standby now sees password spraying, SSH keys added behind your back, and new accounts and persistence on Linux servers. Backup hosts can refuse deletes, so one compromised server cannot wipe everyone's backups. Includes the host agent standby-agentd 0.156.0.

Added

• Password spraying is detected across all your servers. One address trying ten or more usernames in 15 minutes, or one username failing from eight or more addresses in 30 minutes, is reported on the threat feed. A password login that then succeeds as one of those usernames pages, like any attacker login. Key logins are not affected. Servers that report failed logins only through fail2ban do not name the username, so the check cannot see them.
• SSH keys added behind your back are reported. The agent watches every authorized_keys file it can read and reports keys that appear or go. A key Standby did not add pages, and a login with it is critical from any address.
• New accounts and persistence on Linux servers are reported. New or changed accounts, group membership, sudo rules, cron entries and service units are reported. A new root account, a member added to sudo or wheel, a sudo rule Standby did not write, or a cron line or service that downloads or runs from /tmp reaches the bell. Nothing here pages. Changes made in the minutes after Standby ran a script or installed a package on that server are marked as Standby's.
• Backups can be write-once. A backup host can refuse deletes. Your servers can add backups and read them back but cannot delete them, and the backup host removes old backups itself, daily, by each policy's retention. New backup hosts are write-once; the backup map shows Write-once: yes / no for each one.
• A new alert rule flags unusual outbound traffic. It fires when a server sends far more than it has over the last seven days, for at least 10 minutes and at least 5 MB/s. It records an alert and does not page, and it skips workstations. Alert rules can now have a floor, apply to servers only, or not page.

Fixed

• ATT&CK coverage counts the whole period. On busy installs "last 30 days" only read the newest day or so of events. It now counts every event in the period.
• Web attacks count on the ATT&CK page. Web application firewall events were matched to no technique, so SQL injection and file-access attempts showed zero. They are now counted, and the threat feed shows the technique for each event.

Upgrading

• Update your agents (Fleet → Update all agents). Each server first sends a baseline of its keys and accounts; that baseline does not alert.
• For each existing backup host, open Backups, choose the host and press Make write-once. Press it again after you change a policy's retention.

Why

Most intrusions show up first as a login that should not have worked, a key or account that should not exist, or a backup that is suddenly gone. This release watches for each of them and keeps the alerts quiet unless something is actually wrong.

A problem that keeps coming back pages once instead of every time, and postmortem drafts stay accurate until you edit them.

Fixed

• Flapping stops paging over and over. A server that dropped out and came back every few minutes sent a page each time it returned. The first alert and the first return still page. Another return within two hours is recorded on the incident without paging, and every period still shows on the incident.
• Postmortem drafts follow the incident until someone edits them. An automatic draft no longer freezes the moment it was first opened. Until a person edits it, it is rebuilt from the incident each time, in your timezone, and lists each period when the problem came back.
• Shorter timeline notes. "It happened again after 6 min." and "Every affected host is back." replace long sentences that repeated on every recurrence. Older incidents show the short form too.

Auto-block now lifts blocks the old scoring put on your own addresses, and a single security event no longer looks like an outage that never ends.

Fixed

• Auto-block lifts its own mistakes. An address auto-blocked by the old scoring could stay blocked across the fleet until the block expired, even when it was an operator's. Auto-block now lifts a block it made when the address logs in to your servers and scores under the threshold by today's rules. It uses the same unblock as the IP control page, and the audit log records it. Blocks on addresses that have never logged in are left alone.
• A single security event is not shown as an ongoing outage. A malware hit or a login alert shows when it happened and whether anyone has reviewed it, instead of "still happening, 1d 20h so far".
• Active alerts show the incident's reference, not its internal number.
• Incidents no longer list "its" or "call" as the affected host. Titles such as "stopped answering on its public address" or "Nobody is on call" no longer produce a fake host name.
• The dashboard says it is loading instead of "No incidents yet."

Your team's own logins stop setting off alarms, and every incident now shows when the problem actually started, when it ended and how long it lasted.

Added

• SSH keys are judged by the key, not the address. This takes effect with the next host agent update. A login with a key Standby knows stays quiet from any address. A key that appears in authorized_keys without Standby adding it raises a page, and so does a login with such a key. Review or adopt the key on Access keys and it stops paging.

Changed

• Active alerts points to firing incidents it cannot list. Incidents raised straight from a security event have no alert row. The page now says how many there are and links to them.

Fixed

• Standby no longer bans and pages its own operators. A web login prompt counted as a failed SSH login, every successful login raised the address's reputation, and bans Standby had pushed were counted again as new evidence. Together these could block an operator's address across the fleet and page on every login after that. Only real login failures count now. A successful login never raises a score. Standby's own bans are not evidence. Addresses that logged in during the last 30 days are rescored when you update.
• One incident per login, not three. A graded login no longer also opens extra incidents titled with the raw event name.
• Incidents show real dates and times. The incident list shows Started, Ended and Lasted, with a date and time in your timezone. When a problem came back several times, it says how many and lists each period, so a site that went down four times for ten minutes no longer reads as one four-hour outage. Active alerts show when each one started and how long it has been firing. Postmortems use your timezone.
• The activity timeline no longer blames Standby's own work. Bans, scans and checks Standby ran automatically are folded into one line. They are no longer counted as likely root causes.

SSH hardening now takes effect on Alpine hosts, and when a setting still overrides it, the host page names the file and line. Updates the host agent to standby-agentd 0.155.0.

Fixed

• Apply hardening reloads sshd on Alpine. It used to stop at "systemctl not found", leaving the hardening installed but not in effect. Hosts without systemd now reload sshd with rc-service. SSH key rotation does the same.
• The hardening card names the setting that wins. When a setting such as PermitRootLogin yes still applies after hardening, the card shows the file and line it comes from, including lines in the main sshd_config above its Include, and says when sshd never reads the drop-in directory. Files the agent cannot read are listed rather than skipped, and the card refreshes after a failed apply as well as a successful one.

Upgrading

Update your agents (Fleet → Update all agents), then press Apply hardening again on hosts where it stopped at "systemctl not found". Make sure you can log in with an SSH key first: hardening turns password login off.

Alpine hosts are read honestly: failed SSH logins are seen again, and patch, time-sync and firewall readings say what was actually measured. Updates the host agent to standby-agentd 0.154.0.

Fixed

• Failed SSH logins on Alpine hosts are seen again. Alpine's system log is readable only by root, so the agent saw no logins and the Security tab showed "0 events" during a live password attack. The agent now reads that one log through a narrow root reader, follows it across rotation, and records failed passwords and unknown users as authentication failures, one row per attacking address. Hosts without fail2ban on other distributions now report failures too.
• A host whose logins Standby cannot see now says so. The Security tab shows a red notice with the reason instead of an empty list that looks like a clean bill of health, and the Threats page links every such host.
• The Updates tab no longer shows "Security 0" on Alpine. Alpine does not say which updates are security fixes on the host, so the count now reads unknown — not measured.
• Time sync and the firewall are measured on Alpine. Time sync is read from the kernel clock state where timedatectl and chronyc cannot answer (busybox ntpd reads as unmeasured, never as not synced). The nftables posture is read on every heartbeat, and the Firewall tab header says what was read rather than always claiming a managed default-deny.
• Compliance wording and capabilities match the host. Automatic security updates are worded for the host's platform. Linux hosts stop advertising Windows Update, Windows Firewall and Remote Desktop; virtual machines stop advertising BMC, locate LED and SMART; SNMP and PDU actions need the net-snmp tools. An antivirus that can be installed but is not is shown as ClamAV (can install).
• Alpine package vulnerabilities no longer name an upstream release as the fix. The finding stays; the row says the fix is an Alpine package update (apk upgrade).
• nginx without a status page is no longer polled every 30 seconds. The agent remembers the miss for an hour, stops reporting a running nginx as down, and the Service console says nginx status endpoint not configured. Enabling stub_status now works on Alpine.

Security

• Twilio settings require a same-origin request. The cross-site request check exempted the whole Twilio address range, including the signed-in settings, test SMS and test call actions. Only Twilio's own call-backs are exempt now.

Upgrading

Update your agents (Fleet → Update all agents) to get the new Alpine readings and login detection.

Telnyx joins Bird and Twilio as an SMS and voice provider for paging, and every carrier is now set up in one place: Integrations.

Added

• Telnyx can now carry your SMS and voice pages. Open Integrations → Telnyx, enter the API key, the account public key, the SMS sender, the voice caller ID and the TeXML application ID, then send a test SMS or place a test call from the card. Choose Telnyx for SMS, voice or both on Settings → Telephony.
• Telnyx calls take the acknowledgement in the call. The call reads the incident aloud; pressing 1 acknowledges it. SMS pages can be acknowledged by replying ACK <reference>, and Standby texts back to confirm.

Changed

• Twilio is now configured under Integrations, next to Bird and Telnyx. The Twilio card, with its test SMS and test call buttons, moved from Settings → Telephony to Integrations → Twilio. Telephony keeps the provider choice and links to each card. Saved Twilio settings are unchanged.

Upgrading

To use Telnyx: in the Telnyx portal, set your messaging profile's inbound webhook to https://<your domain>/api/telnyx/events and assign your voice number to a TeXML application. Without the account public key, Telnyx SMS replies and delivery receipts are refused.