The Helious Status Page: Uptime You Can Check Yourself

Helious Desk · 25 October 2026 · 5 min read

A news terminal is only useful if it is there when the number prints. From today you can see for yourself whether Helious is. The status page at helious.io/status shows the state of every surface we run, a 90 day uptime figure for each, and every incident, open or closed, with the times. Here is what is on it, how the figures are measured, and what we have promised to hit.

1. What is on the page?

Seven rows, one per surface a customer touches.

• Terminal. The screen at helious.io loads and serves a page.
• REST API. The public API answers a keyless request with a proper reply.
• Live stream. The websocket opens and sends its first frame.
• MCP server. The address an AI agent connects to lists its tools.
• Market data and news pipeline. The worker that scores releases and reads auctions is publishing.
• Ask Helious. The model behind Ask answers a credentialed call.
• Billing and renewals. The biller that runs renewals has checked in recently.

Each row carries its current state, a bar of 90 cells, one per day, and its uptime over those 90 days beside the target we set for it. Green is operational, amber is degraded, red is an outage, and grey is a day with no readings, which we show rather than hide.

2. How is it measured?

A monitor probes each surface once a minute over the public internet, through the same addresses a customer uses. It loads the terminal, calls the API, opens the stream and waits for the first frame, and asks the MCP server for its tool list. The pipeline is read from the same health light our hosting uses to route traffic. Ask Helious is a real call to the model provider that costs no tokens. Billing is the renewal biller's heartbeat.

Every probe has a ten second limit. A probe that cannot judge, for example because a bot challenge got in the way, writes nothing for that minute and never counts against uptime. Every reading is kept, and the daily totals are kept for good.

3. How is the uptime number worked out?

Minutes that were checked and were not down, divided by minutes that were checked. A minute the monitor itself did not run is not in the denominator. Degraded minutes count as up, because the surface answered, slowly or partially. They are shown on the bar and in the detail, not in the figure.

A day on the bar is painted by its worst stretch, with a floor so one lost probe does not paint a day red: red from five down minutes, amber from one down minute or five degraded minutes, green otherwise.

4. What are the targets?

99.9% on the four surfaces you touch directly: the terminal, the REST API, the live stream and the MCP server. 99.5% on the three behind them: the pipeline, Ask Helious and billing.

Those are the numbers this design can meet and this page can prove, so they are the ones we publish. 99.9% is about 8 hours and 46 minutes a year. We will review the targets once there are 90 days of data on the page and raise one only after the measured figure has beaten it for a full quarter. We did not write a bigger number because it reads well.

5. Who opens an incident?

Three parties, and the page says which.

• The monitor. Three consecutive failed probes on a surface open an incident, and three consecutive clean ones close it. An outage pages the team at once. An incident the monitor opened carries a small "opened by the monitor" tag.
• A second check from outside. Every ten minutes a check runs from machines that are not ours. It asks the public endpoints the questions a visitor's browser asks and looks for the signature of a real failure: an empty calendar, a worker that has gone quiet. When it finds one it opens an incident titled "Outside check" on the page, and it closes it when the site is serving data again. It is there for the case where the monitor inside cannot see the problem.
• A person. We open incidents by hand for a planned deploy window, a partial fault the probes cannot see, or a problem at a provider. Those carry no monitor tag.

6. What happens if the status page itself goes down?

The page is served from the edge, not from the API host, so it stays up whenever the edge does. It fetches the live readings from the API with a six second limit and keeps a copy of the last good answer. If the API host does not answer, the page shows that last copy under a banner saying so, with the REST API row set to outage and every other row marked as an old reading. That is the honest state for a status page that is not hosted by a third party, and it is why the outside check exists.

7. Can my code read it?

Yes. The same data is at api.helious.io/api/status as JSON, public, no key, cached for 30 seconds. The overall state, each component with its 90 day and 30 day figures and daily cells, and the incident list are all in it. There is also a plain text twin of the page at helious.io/status.md for an agent that reads Markdown.

Where to go next

• The status page. Bookmark it beside the terminal.
• The developers page, for the API, the stream and the MCP server the page is watching.
• The live feed, free to read.