What changed, and when

2026-10-04 · HopMango in four sections

The portal was rebuilt around one idea: simple for people, complete for assistants and operators. Nothing you configured was deleted, and nothing was removed from the API.

• Four sections: Monitored, Problems, Team, Settings. Monitored is one list of your servers, websites and scheduled tasks, with what is failing first. Problems says what failed, since when, who is on it and who is called next. Team is who is on call. Settings holds your account, plan, API keys, connections, status page, organizations, the audit record and Mango AI.
• Everything is still on screen. A server's page shows all that is monitored on it, the services found on it with Monitor or Ignore, and its charts over a range you choose. A website shows its results per location, its last failures and its real users. A problem shows its full history.
• On call in one choice. Just me, or a weekly rotation; who is called if they do not answer; Cover me when you are away; and how you are reached, by app, call or e-mail. Whoever is on call is told; if they do not answer in five minutes, they are called; then the backup is called.
• Choose who is on call. Until somebody is chosen in Team, nobody is told, and Team says so plainly instead of describing an order that does not exist.
• Plain actions where you look for them. Silence for an hour, until tomorrow or from and to; I'm on it; Resolved; Call someone else.
• Several destinations. Problems can be posted to more than one Slack, Teams or Discord room and mailed to a member, and every room that heard a problem open hears how it ended.
• One place to ask Mango. The floating assistant is the only conversation, it can start about a problem or a server, and its history can be read and deleted.
• Change your password where you are, with the current one and the new one twice, and leave an organization yourself.
• For assistants: a small first layer of MCP tools mirrors the four sections and their actions, and the full set stays underneath for everything else.
• For operators: the new hm command-line tool reaches every operation of the API, with the everyday actions as short commands. Through the API you can also preview a change to an alert rule before saving it and read one timeline of who is told and when.

Security

• Names and fields of what you monitor no longer accept control characters. A crafted name could have changed the configuration of the agent on a server. We found no such entry in production before the fix.
• A password link issued by an owner is mailed to the person, never shown to the owner, and invitations are mailed the same way.
• Test and verification calls are limited per person, organization and number per day, need a plan with calls, and are refused for premium numbers and unsupported countries.
• A website's secret headers go only to that website. They are not sent to another host, and a redirect elsewhere does not carry them.
• The key of a real-user site can be replaced, with a day in which both work, and the snippet no longer sends page titles or error texts.
• An assistant cannot create API keys, and a person who leaves an organization loses theirs.
• The agent is stricter with what HopMango sends it: nothing runs more often than every ten seconds, internal addresses are guarded for every kind of remote test, and its memory stays bounded when its buffer is full.

2026-10-03 · Connect ChatGPT, and a first destination that works at once

• HopMango is ready for ChatGPT. Connecting an assistant asks for your consent and never grants more than your own role allows. You see the assistants you connected and disconnect them under Settings. Assistants connected before this day connect once more.
• The first destination you add is where alerts go. Critical alerts reach it from that moment, with nothing else to set up; the API can opt out.
• Fixed: long lists no longer skip a row between one page and the next.
• The privacy policy covers connected assistants and the account e-mail they may read.

2026-09-15 · Mango AI answers briefly, and says what it does not know

• Shorter answers about your organization, in your language.
• Missing data is not a cause. No visitors measured yet does not mean a broken snippet, and no room configured does not mean nobody would be told. Mango says which of the two it is.

2026-09-11 · Know what is monitored, and how you will hear about failures

• A new overview brings monitored resources, current incidents and notification setup together. Missing data stays visible instead of being mistaken for good health.
• Ask Mango from any screen. The floating assistant keeps your conversation as you navigate, understands which resource and time window you are viewing, and answers in your portal language.
• Start monitoring with fewer decisions. Server installation prepares its default group; website setup starts with a name and URL. Advanced configuration remains available.
• Understand your alerts. Resource pages explain the timing. Notification setup separates the rule, the routing preview and a real delivery test, so configuration is never sold as proof.
• The same simpler journey through MCP, with coverage, installation, first-route setup and confirmed notification testing. The Skill now opens with a short guide and loads detail by task.
• Existing API integrations and the iOS app retain their contracts.

2026-09-10 · A server is monitored the moment you install the agent

Adding a server used to end in a list of everything the agent could see, each line waiting for a click. Now it ends in a monitored server.

• CPU, memory, load and every real disk are monitored from the first minute, with thresholds that do not wake anybody for nothing. There is nothing to accept and nothing to configure.
• The agent proposes services, not an inventory. It looks for nginx, Apache, PostgreSQL, MySQL, Redis, Docker, OpenSSH and TLS certificates, and offers each one as a single decision: monitor it or ignore it. Individual units, ports and processes are no longer proposed; you add one yourself when you want it.
• Ignore is remembered. A service you ignored stays ignored after the agent restarts. You can restore it from the host page.
• The host page starts with what matters: what is wrong right now, then four charts, CPU, memory, load and all the disks on one, then the services on the host. The internals live under an *Advanced* fold.
• The privacy policy names Amazon Web Services as the operator of the hosted service.

2026-09-08 · Three servers, a public status page, and a digest in the morning

The platform moved to redundant hardware, the portal was rebuilt to be read at a glance, and the API stopped being something you had to have an account to read about.

• HopMango now runs on three servers in three zones with automatic failover. If one of them goes away, the other two keep answering, and nothing you configured is lost.
• A public status page. Choose the monitors you are happy for the world to see and publish them on a page anyone can open, with no account and no password.
• A daily digest by e-mail. One message a day with what happened to your systems, so nobody has to keep a dashboard open to stay informed.
• A failing website check now says why. Instead of a bare “down”, a monitor names the cause it actually found: the certificate, the name that did not resolve, the redirect that went somewhere else, the request that never finished.
• Invite your team by e-mail. Roles decide who can change what, removing somebody takes their access away on their very next request, and every mail we send says when its link stops working and arrives in the language of the person reading it.
• The portal, rebuilt. Hosts, checks, events and settings were reworked to be understood in one look, the whole thing works on a phone, and the first hour after signing up now explains itself instead of leaving you on an empty screen. Every sentence is available in English and in Spanish, including the ones the platform writes itself.
• Prices on the website, and the door open. The plans and what they cost are published, and anyone can open an organization on the free plan without a call and without a card.
• Thirty days of Team, on the house. Every new organization starts on the Team plan for thirty days: more servers, more website checks, a month of history and alerts by phone, with no card asked for. When the month is up the organization goes on with the free plan and nothing is deleted; whatever is over the free limits is simply frozen until you choose a plan.
• And the trial tells you what its ending will change, before it ends. The billing screen lists exactly what is different for your organization on that date - which of your things go over the free limits, which ways of alerting stop, and how much history is kept - and if nothing of yours changes it says so in one line. We also write, a week, three days and one day before, so the date is never a surprise.
• Public documentation. The API and the way to connect an assistant to HopMango are now documented in the open, in both languages, next to a page that says plainly what our agent will never do on your machines.
• Six Skills you can download. An assistant you already use can learn HopMango from them and work with your monitoring the way you would.
• The installer verifies signatures by default. What it downloads is checked against our signature before a single file is installed, and it refuses to continue if the check fails.
• The agent reloads its file. Edit its configuration file and it picks the change up without a restart, and it will now answer questions about its own state from the shell.
• A PostgreSQL check. It watches archiving, the write-ahead log and replication, and tells you when a database has quietly stopped keeping up.
• Mango AI, more useful and still only a proposal. It names the thing it is talking about, groups incidents that share a cause and shows what it spent in dollars. It proposes; it never touches your systems.

2026-09-07 · On-call, Mango AI, three regions, and a way to sign up

The night the product stopped being a monitoring agent and became something a team can be on call with.

• Mango On-Call. Rotations that survive daylight saving time, escalation ladders, covering a colleague's shift, and acknowledging a page from a link without signing in first. Voice calls go out through our own number.
• Mango AI. It summarises an incident, suggests a likely cause and answers questions about what it is allowed to read. It proposes; it does not act, and it never sees your secrets.
• Website checks run from three regions. A site is watched from three points of presence, so one bad network between us and you is not reported as your outage.
• A site that refuses our probes reads “blocked by the target”, not “down”. And a monitor can be sent through a firewall with a header of your choosing, so a protected site can be watched honestly.
• Sign up with an e-mail address. A visitor becomes an organization by verifying an address. No sales call, no waiting list.
• Plans, and limits that are enforced everywhere. Every limit bites in the platform rather than in the screen, so an API key and an assistant are refused exactly as a browser is, and the refusal names the plan, the current count and the call that lifts it.
• The portal speaks Spanish. Every screen, in English and in Spanish.
• Silence an incident from its own page, with or without an account, reopen one that was closed too early, and see the checks an incident was built from.
• Delete your own account from the portal. Without asking us, and without a form.
• Your visitors' countries. Real-user monitoring now reports where a session came from.