What shipped, in the order it shipped.

A verifiable record of every incident, and less work to get there

This release is about the record behind your status page: the measured impact of every incident, evidence you can hand to customers and auditors, and easier ways to bring in what you already run.

• Incidents carry their measured impact: when it started and ended, what and who was affected, error budget spent, and which checks saw it.
• AI-drafted updates are written from that measured impact and checked against it before you see them. Nothing is published without you.
• Bring the alerts you already have: Prometheus Alertmanager, Grafana, Datadog, CloudWatch and any JSON source can drive status and incidents.
• Test a check before you save it, through your own agent.
• Business plans: a signed evidence record for any incident, a private reliability portal for each customer, and audit log export.
• Twelve-month uptime and incident reports, as PDF or CSV.
• Connect Claude, Cursor and other AI tools by signing in, with no key to copy. Other tools can add their findings to an incident as internal notes.
• Move from Atlassian Statuspage: components, incident history, maintenances and consenting subscribers, without re-sending past notifications.

Faster, sturdier status pages your visitors can follow their way

Status pages now hold up better under load and outages, and give visitors more ways to stay informed.

• Subscribe to only the services you care about, by email, Slack, Microsoft Teams or a signed webhook.
• Upcoming maintenance on the page, plus a calendar feed you can add to any calendar app.
• Status pages in eight languages, set per page.
• A permanent link for every incident and maintenance, used by feeds and emails.
• Search settings in the page editor: a custom description, or hide a page from search engines.
• Faster loading, and pages keep serving their last known status during an outage on our side.
• Accessibility improvements: screen reader announcements, keyboard access to daily uptime, and a chart legend.
• Fixes to subscribing and unsubscribing, access-restricted pages and embeds.

A redesigned Observer, from sign-up to status page

A broad release across the product: a new look for the console and status pages, a quicker path to your first live page, and a long list of refinements and fixes.

• Improved console UI: navigation, tables, charts, forms and dialogs.
• A new status page layout, with refreshed themes and subscriptions.
• Simpler onboarding, from sign-up to a published page.
• Redesigned emails and sign-in.
• Agent and probe improvements, with faster pickup of changes.
• Set which customers an incident or maintenance affects through the public API and MCP server.
• Config apply now warns about sections it doesn't manage instead of ignoring them.
• Heartbeat monitoring for cron jobs, backups and workers, directly or through an agent.
• Reliability Center: a public page with your commitments, attainment, uptime and incidents, plus private customer links to their contract and signed statements.
• SLA Ledger for per-customer contracts, credits and approved monthly statements.
• Planned maintenance pauses SLO error budgets by default, with a per-SLO option to count it.
• Fixes across alerts, uptime figures and subscriptions.

Metric alerts group instead of flooding

Metric alerts on messaging channels now group: several metrics changing status within a minute arrive as one message instead of one per metric, and when an agent goes offline its metrics' "stopped reporting" alerts fold into the agent alert instead of flooding the channel. On-call channels (PagerDuty, Jira, Splunk, Grafana IRM) keep per-metric alerts so breach→recovery still auto-resolves.

Connect an AI agent in one dialog

The API keys page now has a guided setup for Observer's MCP server: create a scoped key, copy a ready-made config for Claude Code, Claude Desktop, Cursor, VS Code, or any MCP client, optionally add the Observer skills, and verify the connection. Read-only by default; an assistant can only do what the key's scopes allow.

Route alerts the way your team works

Each alert destination now has routing controls: a minimum-severity floor, a choice of which metric flips alert, and a daily-digest mode for low-urgency channels. Recoveries always deliver, so an alert that pages you also closes itself. You can snooze a destination for 1 to 24 hours, see its 7-day delivery success rate, expand any delivery to inspect exactly what was sent, and re-queue a failed delivery with one click. Metric alerts are suppressed during scheduled maintenance on the affected pages: visibly marked as skipped, never lost.

Three-state SLOs, evidence-pack previews, a fuller executive view, and built-in support

SLOs now distinguish observed downtime from monitoring silence. A per-SLO policy decides what unobserved periods mean: count against the budget (the default, unchanged), drop out of the math with monitoring coverage disclosed alongside the number, or count as up where a contract says so. The same answer applies everywhere at once: the live error budget, the SLO page, SLA reports, and compliance packs. And when something was telemetry rather than downtime, you can exclude the window with a written reason; each exclusion is capped at 30 days, audit-logged, and listed in evidence packs, so numbers never improve silently.

• SLO windows start when the SLO is created, so a new SLO no longer inherits history from before it existed; exhausted budgets now show a projected recovery date
• Compliance packs: preview any pack inline, dry-run one before generating, per-SLO monitoring coverage, and maintenance plus adjudicated exclusions listed with reasons
• Executive dashboard: error-budget trajectory for your riskiest SLOs, incidents per month by severity with an MTTR trend, and subscriber and customer growth
• Support, built in: file a request from the console with opt-in diagnostics (screenshot, recent logs, failed requests, optional screen recording); attachments are stored encrypted and deleted after 60 days; track requests and revoke access grants under Settings → Support
• unknown_policy is settable via config-as-code and readable from the API

Metric descriptions, SLOs on every layout, and SLOs in the API

Every metric definition has always carried a description; now the page builder can show it. A toggle on the metrics section renders each description as a quiet line under the metric name, on every layout. Public SLOs also stopped playing favourites: the Aurora and Headline layouts now render the SLO cards the other layouts already had, and status.json carries the same SLOs so a dashboard or procurement script can read your targets and remaining error budget directly.

• Descriptions are off by default; the builder warns you they become public
• New SLOs report "collecting" until they have a real day of history, instead of a meaningless number

Builder changes land faster, and can't quietly collide

Saving in the page builder now clears the edge cache for your public page, so a reorder or a renamed group shows up on the next refresh instead of a few minutes later. The builder also refuses to save two metrics that would display under the same public name, which used to quietly merge their uptime histories into one strip.

Agent 1.5.0: batched delivery and build provenance

The agent now ships its queue in batches of up to one hundred samples per request, so a backlog after an outage drains in seconds rather than minutes. Each agent also reports where its build came from: the dashboard shows an Official badge when the running code matches the published release, Source for checkouts, and Modified when an official image has been changed. An update chip appears whenever an agent is behind the latest release.

• One bad sample no longer costs the rest of its batch
• A stuck ingestion sample is retried with backoff and set aside for review instead of being lost

Live status badges for your README

Every public page now serves badge.svg: a live status badge for READMEs, wikis, and dashboards. Three styles match what shields.io users expect, an uptime variant shows the 30-day number, and the badge colour always follows the current status so a red page can never hide behind a green number.

Status in your pocket: Apple Wallet passes

Subscribers can now keep a status page in Apple Wallet. The pass shows the current status, the latest update, and how many incidents are open. When something changes, the pass changes with it and a line appears on the lock screen. Operators can preview the pass in the console and add a test copy to their own phone before sharing it.

• The card takes the colour of your current status, maintenance included
• Your logo rides along, and a QR code on the pass links back to the page

Find anything: filters across every list

Metrics, agents, pages, and the audit log share a new filter bar. Combine filters, and the view lives in the address bar, so a filtered list can be bookmarked or sent to a teammate and it opens exactly as you left it. The lists themselves now answer more at a glance: when each check last reported, how many people subscribe to a page, and whether a custom domain is live.

Maintenance windows, without the sharp edges

Scheduled maintenance can be canceled without deleting it. The history stays, the public page gets a short note, and subscribers are told. The updates list also warns you when a window was supposed to start but did not, and the forms no longer accept a window that ends before it begins.

One status, everywhere your visitors look

The main page, the incident history, the embeddable widget, the RSS feed, and the operator preview now all agree on the overall status. Long-running incidents stay in the headline until they are actually resolved, and feed readers see the current status alongside the timeline.

A silent agent can no longer look healthy

When an agent stops reporting, its checks now show as awaiting data everywhere: the dashboard, service targets, status pages, and email reports. Before, a check could hold its last good reading indefinitely and quietly go green through an outage of the thing watching it.

A demo you can run in one command

One compose file starts the agent alongside a small fleet of real services: databases, an API, log and telemetry pipelines. Every kind of check Observer supports runs against something genuine, and a live status page fills in as it happens. A good way to see the product working before pointing it at your own stack.

Watch far more than Prometheus

The agent grew a family of new checks: call a gRPC service, hold a WebSocket open, run a read-only query against Postgres, MySQL, Redis, or MongoDB, read a metric from CloudWatch, Loki, or Elasticsearch, accept pushed telemetry, or register a check of your own. Whatever you watch, credentials stay on your side; only a reading and its verdict leave.

Your monitoring, written down

The Observer CLI turns a workspace into a YAML file and a YAML file into a workspace. Describe your checks, services, targets, and pages once, keep the file in version control, and apply it whenever it changes. A dry run shows exactly what would change before anything does, and objects managed this way are labelled in the console so no one edits them by hand.

Know how your agents are doing

Each agent now reports on itself: how far behind its queue is, how much of the last day it was up, how often it restarted, and whether two copies are accidentally running with the same key. The dashboard turns those into plain chips with a warning when something needs a look.

Status pages on your own domain

A status page can live at an address you own, with certificates handled for you. The console walks through the one DNS record to add and shows where things stand until the page is live.