About down.guide

A free outage feed for people and agents. It reads the official status pages of 406 cloud, SaaS and developer providers every minute, normalizes their incidents, and labels its own evidence (visitor reports) as its own.

Each provider's own status page (Atlassian Statuspage, incident.io, Instatus, Better Stack, status.io, and the AWS, Google Cloud, Slack and Salesforce feeds, with RSS as a fallback). Their words are quoted verbatim and linked. down.guide's own evidence is always labeled as ours, with numbers.

One GET per provider status page per minute, with the User-Agent down.guide and a link to this page, and a conditional request (ETag or Last-Modified) so an unchanged page costs a 304. No credentials.

It adds one to an hourly count for that provider. Reports are only ever shown in aggregate, deduplicated per visitor every 10 minutes, capped per hour, and weighed against the provider's own normal rate so one network can't make a spike. We keep a daily-salted hash, not your IP.

A 0-100 score for every confirmed incident from impact, breadth, duration, the provider's reach and how long it took to acknowledge. The weekly table lists the worst. The methodology page has the exact weights.

From each provider's own incidents, per component, the way Statuspage computes its own: a major outage counts as full downtime and a partial outage as 30%; degraded performance and maintenance don't count. An incident counts against the components it names, and one that names none counts as a single component unless the page lists no components at all. An incident whose updates or components name only cloud regions, under a title that doesn't (us-east-1, europe-west2), is a regional outage: it counts as a partial outage at most, and only for that share of the provider's regions, so one AWS region down isn't all of AWS down. Minor or regional issues (below) never count. The window starts at the provider's first data point.

Incidents confined to a slice of a provider rather than the provider itself: at any provider, a title naming one country, continent or market region ("EU", "US Region", "APAC"), one or two cloud region codes in any case ("us-west-2", "US-West-2"), one data center, cluster, city, point of presence or named instance ("EUR-IS-3", "AUS4S, AUS6S"), or a named subset of customers ("a small number of users", "some merchants", "FedRAMP customers") or of networks ("some Australian ISPs", "certain carriers"; also an ISP or carrier incident whose only region is one country); an incident the provider rates as no impact, with every component operational, is a notice and listed the same way; one carrier route or local market at a messaging or voice API (Twilio, Sinch, Bandwidth, Plivo, Telnyx); one carrier at a shipping API (Shippo, EasyPost); one chain or network at a blockchain node API (Alchemy, Infura); and one asset, network or funding method at a crypto exchange (Kraken, Coinbase). They're listed on the provider's page but never set its state, and they stay out of the open-incident count, the Down Index, the major filter and uptime. A title about all regions, countries or destinations, global or worldwide, all assets, trading, sign-in or the dashboard is a normal incident; one naming three or more region codes is several regions, so a normal incident.

Some providers title an incident with only a ticket number (Snowflake's "INC20000150", Databricks' "ES-2342108"). Where that would be the headline, down.guide shows the components the incident names ("Virtual Warehouses, Snowpark Container Services affected"), or a short line from its summary, instead. The API's title is that readable headline and sourceTitle is the provider's own title, unchanged.

Some catalog entries read the same status page (GitHub Actions and GitHub, SendGrid and Twilio), and some pages cross-post each other's incidents (Akamai and Linode). Each provider page shows it, but sitewide numbers (open incidents, Happening now, downtime, the Down Index) count it once.

An MCP server at https://mcp.down.guide/mcp (also /mcp on this site), a REST API under /api/v1 with the same JSON as each MCP tool, RSS/Atom/JSON feeds you can filter by provider, category, impact, region or your stack, and a .md copy of every page.

Subscribe to a feed; there is no account or webhook to set up. https://down.guide/feed has every incident, /feed?provider=<slug> one provider, and the stack checker builds one feed for all of a project's providers. In Slack, install the RSS app and run /feed subscribe <feed URL> in a channel; in Discord, give the URL to a feed bot such as MonitoRSS. Add format=atom or format=json for Atom or JSON Feed. For scripts, GET https://down.guide/api/v1/incidents takes the same filters and pages by 30 (pass next_cursor back as cursor). To get them by email instead, use the Email tab of the subscribe form on the homepage: Instant for one or more providers (a new or worsening incident, at most one email every 15 minutes), or Daily or Weekly at 9:30am Pacific.

Your email address, the providers and filters you chose, how often you want them, and when we last emailed you. Nothing else: no name, no account, no tracking pixels. A subscription starts only after you click the confirmation link. Every email has a one-click unsubscribe link (also in the List-Unsubscribe header) that removes that subscription. The manage page can unsubscribe you from everything, which keeps just your address, marked as unsubscribed, until you subscribe and confirm again; to erase that record too, send feedback from any page.

Nothing, and there's no account to make. There is no token or coin, and there never will be.

Sazabi, which builds AI-native observability and security. down.guide is a free Sazabi Labs project, like cve.now. It uses no Sazabi customer data.

Open an issue at https://github.com/sazabi/down.guide: it has forms for wrong data and missing providers, and documents how the data is collected. For anything private, use send feedback (the form on any page, the send_feedback MCP tool, or POST /api/v1/feedback). A person reads every note.