Reference
Full page layouts, each shown in the mode that suits it. They all use the same tokens and components. Only the spacing, density and size of display type change from one to the next.
01 / App shell · industrial
A 48px top bar and a 256px sidebar, each separated by a 1px line. Compact density, mono table headers that stick as you scroll, and status shown as a square plus a word.
6 services across 3 regions
Requests / 24h
2.41M
p95 latency
88ms
Error rate
0.04%
Spend / mo
$1,284
| Service | Region | Status | p95 | Requests |
|---|---|---|---|---|
| edge-router | us-east-1 | Healthy | 41 ms | 2,410,882 |
| api-gateway | eu-west-2 | Healthy | 88 ms | 1,092,441 |
| image-proxy | ap-south-1 | Degraded | 612 ms | 318,204 |
| webhook-fanout | us-east-1 | Failing | — | 0 |
| cron-worker | eu-west-2 | Paused | — | 0 |
Scroll sideways to see the whole preview at desktop width
02 / Marketing · editorial
A two-column hero with a 1px line between the columns, a feature grid with shared borders so the section reads as one table, and a single inverted call-to-action band.
Infrastructure
Ship to the edge in one command
310 locations, immutable builds, instant rollback. No YAML, no control plane to babysit.
No credit card · 4,000 free builds
02 / Platform
Everything the team needs
Every deploy is content-addressed. Roll back in under a second.
Requests terminate at the nearest of 310 locations, with failover.
Certificates issued, renewed and stapled with no configuration.
Start shipping in under five minutes
Free for personal projects.
Scroll sideways to see the whole preview at desktop width
03 / Documentation
Three columns separated by 1px lines. Body text is capped at 68ch, code blocks have a mono strip naming the language, and the active heading gets an accent line.
Guides / Routing
Requests are matched against your route table in order. The first match wins, so put specific patterns above catch-alls.
Defining routes
Routes live in orbit.toml at the root of your project.
[[routes]] pattern ="/api/*"service ="api-gateway"timeout =30[[routes]] pattern ="/*"service ="edge-router"
Order matters
A catch-all above a specific pattern will shadow it entirely.
Scroll sideways to see the whole preview at desktop width
04 / Flows
A 400px sign-in card on a darker background with the grid showing, and a settings page where each row is a cell in one shared grid.
Account / Settings
General
Auto-scale
Add instances when p95 exceeds 200ms.
Preview deploys
Build every pull request.
Region
Primary write region.
Danger zone
Delete project
Permanent. Requires typing the project name.
05 / AI agent · blueprint
Sessions on the left, the transcript in the middle, and the agent's plan and changed files on the right. Tool calls are log lines you can expand. The approval request is the only thing in the warning colour. Type in the composer and it streams back a reply.
send.test.ts fails about one run in five on CI but never locally. Find the cause and fix it; do not touch the mock server.
CI-only and intermittent points to timing. CI boots the mock in the same job, so the first request can land before it is ready. If send() has no retry, a single 503 fails the test.
Check the sender first, then reproduce with a cold mock.
src/webhooks/send.ts:4 export async function send(hook, body) src/webhooks/queue.ts:31 await send(hook, payload)
export async function send(hook: Hook, body: string) {
const res = await fetch(hook.url, { method: "POST", body });
if (!res.ok) throw new WebhookError(hook.id);
return res;
}{ "command": "MOCK_COLD=1 npm test -- webhooks", "timeout": 60 }Output · exit 1FAIL webhooks/send.test.ts
✗ delivers to a cold endpoint
WebhookError: hook_01H8 (503 Service Unavailable)
Tests: 1 failed, 11 passed+ for (let attempt = 0; attempt < 5; attempt++) {
+ const res = await fetch(hook.url, { method: "POST", body });
+ if (res.ok) return res;
+ await sleep(200 * 2 ** attempt);
+ }
- if (!res.ok) throw new WebhookError(hook.id);Reproduced it. A cold mock returns 503 for roughly 300ms1, and send() throws on the first non-2xx response. I added a bounded retry:
export async function send(hook: Hook, body: string) { const res = await fetch(hook.url, { method: "POST", body }); if (!res.ok) throw new WebhookError(hook.id); for (let attempt = 0; attempt < 5; attempt++) { const res = await fetch(hook.url, { method: "POST", body }); if (res.ok) return res; await sleep(200 * 2 ** attempt); } throw new WebhookError(hook.id);}Run the full suite 20 times against a cold mock to confirm the fix:
for i in $(seq 20); do MOCK_COLD=1 npm test -- webhooks || exit 1; done
Scroll sideways to see the whole preview at desktop width
06 / Marketing · editorial
A feature grid with one wider cell, then pricing laid out as a plain table. The recommended plan gets an inverted header instead of a bigger card. The billing toggle works.
01 / Platform
Everything between git push and production
Every build is immutable and content-addressed. Rolling back is a pointer move that lands in under a second, at every edge location at once.
Live
Rolled back in 0.8s
Superseded
Requests terminate at the nearest location with automatic failover.
Certificates issued, renewed and stapled without a line of config.
p50, p95 and error rate per route, kept for 30 days.
Isolates, not containers. Nothing to keep warm.
02 / Pricing
Pay for what you ship
Hobby
$0 /month
Personal projects and experiments.
Team
Recommended$20 /month
Per member. For teams shipping weekly.
Enterprise
Custom
Volume pricing, SLAs and support.
Compare plans
| Feature | Hobby | Team | Enterprise |
|---|---|---|---|
| Deploy | |||
| Build minutes / month | 100 | 6,000 | Custom |
| Concurrent builds | 1 | 4 | 24 |
| Instant rollback | |||
| Security | |||
| Audit log | 30 days | 1 year | |
| SAML SSO | |||
| Uptime SLA | 99.9% | 99.99% | |
Scroll sideways to see the whole preview at desktop width
07 / Support · editorial
A heading column that stays in place as you scroll, next to a numbered accordion that opens one answer at a time. A segmented control filters the questions by category. The section ends with a real person to contact, so readers who don't find their answer still have somewhere to go.
03 / FAQ
Questions, answered plainly
Can’t find it here? The docs go deeper, and a human answers the rest.
Still stuck?
Median first reply: 38 minutes
Wall-clock time from the first build step to the upload finishing, rounded up to the next minute per build. Cached steps that are skipped cost nothing.
Nothing stops. You get an email at 80% and 100%, and overage is billed at the published rate:
Yes, on every plan. Point a CNAME at edge.orbit.app; certificates are issued within a minute and renew on their own.
Node 20 and 22, Bun, Deno and any static output. Set it per project in orbit.toml.
Build artefacts live in the region you pick at project creation. Logs are kept for 30 days and never leave that region.
SAML SSO and SCIM provisioning are on the Enterprise plan. Google and GitHub sign-in are on every plan.
6 questions
Scroll sideways to see the whole preview at desktop width
08 / Content · editorial
The index puts one featured post at the top and the rest in a grid that reads like a table of contents. The article keeps text to 68ch, has a sticky table of contents beside a vertical line, and uses every content component: lists, a blockquote, code, a callout and a figure with a caption.
Blog
Engineering notes
Four million scheduled jobs a day, one migration weekend, and the three invariants that let us do it with zero missed runs.
Why we alert on the whole distribution instead of one percentile.
Push anything and get a URL. Here is what it costs us, and why it is free.
Per-route timeouts and retries, a faster CLI and 11 fixes.
A single percentile hides the shape of the distribution. Here is what we alert on instead, and the incident that changed our minds.
One Tuesday in July our p95 latency looked perfect: 88ms, flat all afternoon. Meanwhile 4% of checkout requests were timing out. The dashboard was right about p95, but p95 was the wrong thing to be watching.
A percentile is one point on a curve. Two very different services can share a p95:
We were not measuring the service. We were measuring one number the service happened to produce.
We now keep a full histogram per route and alert on the share of requests over budget, which is closer to what a user feels. The query is short:
SELECT route, count_if(latency_ms > budget_ms) / count(*) AS over_budgetFROM requestsWHERE ts > now() - INTERVAL '5 minutes'GROUP BY routeHAVING over_budget > 0.01;Budgets are per route
Checkout gets 300ms, image resizing gets 2s. One global budget recreates the p95 problem.
Scroll sideways to see the previews at desktop width
09 / Analytics · industrial
Filters sit in one row above everything they affect, with the date range first. Below them are KPI tiles with sparklines, a line chart with two series (the comparison line is gray and dashed, so it still reads in greyscale), a bar chart labelled on the bars, and a team table with workload meters and row menus. Hover over the charts or step through them with the arrow keys.
Insights
Requests
14.2M
▲ 12.4% vs prev 7d
p95 latency
88ms
▼ 4ms vs prev 7d
Error rate
0.12%
▲ 0.03pp vs prev 7d
Spend
$1,284
64% of $2,000 budget
| Day | This period | Previous |
|---|---|---|
| Sep 19 | 1.8 | 1.6 |
| Sep 20 | 1.6 | 1.5 |
| Sep 21 | 1.7 | 1.4 |
| Sep 22 | 2.1 | 1.8 |
| Sep 23 | 2.3 | 1.9 |
| Sep 24 | 2.2 | 2.0 |
| Sep 25 | 2.5 | 1.9 |
| us-east | 5.8 |
|---|---|
| eu-west | 4.1 |
| ap-south | 2.2 |
| us-west | 1.5 |
| sa-east | 0.6 |
| Member | Role | Workload | Pages 7d | Status | Actions |
|---|---|---|---|---|---|
MKMina Kato | Lead | 3 | On call | ||
JRJo Rivera | Admin | 11 | Overloaded | ||
AOAda Okafor | Member | 7 | High | ||
SLSam Lee | Member | 1 | Off rota |
Error rate above 1%
Paged Jo Rivera
Deploy succeeded
Sam Lee joined the rota
Budget alert at 60%
Scroll sideways to see the whole preview at desktop width
10 / Documentation · blueprint
Sidebar sections have mono headings on darker strips, and groups open three levels deep along drawn tree lines. The article has a breadcrumb, a tab for each package manager, a parameter table, an accordion for options you rarely need, and previous and next links. The version switcher is a menu.
A rewrite serves a different path than the one requested, without changing the URL the visitor sees.
Install the SDK if you want to define rewrites in code rather than in orbit.toml:
npm install @orbit/sdkpnpm add @orbit/sdkbun add @orbit/sdkPatterns may use named groups. Each group is available in the destination as :name.
[[rewrites]]source = "/blog/(?<slug>[a-z0-9-]+)"destination = "/posts/:slug.html"status = 200| Key | Type | Default | Description |
|---|---|---|---|
| source | string | — | Path or regex to match. Required. |
| destination | string | — | Path to serve instead. Required. |
| status | number | 200 | Response status to send. |
Rewrites run after redirects
A redirect that matches first will win. Check matching order.
Set destination to a full URL. Orbit proxies the request and streams the response back; caching headers from the origin are respected.
Query strings are forwarded by default. Set query = "drop" to strip them.
Scroll sideways to see the whole preview at desktop width
11 / Storage · industrial
A folder tree, a breadcrumb that doubles as the path, and a table you can right-click. The right-click menu repeats the actions in each row's overflow menu, which stays visible. Drop files on the dashed area to add them to the upload tray.
| Name ▲ | Type | Size | Modified | Actions | |
|---|---|---|---|---|---|
fonts/ | Folder | — | Sep 21 | ||
img/ | Folder | — | Sep 24 | ||
app.4e2c.css | text/css | 48.2 KB | 12:04 | ||
app.91bf.js | text/javascript | 212.6 KB | 12:04 | ||
hero.avif | image/avif | 1.4 MB | Sep 19 | ||
robots.txt | text/plain | 112 B | Aug 02 |
logo-mark.svg
inter-var.woff2
Scroll sideways to see the whole preview at desktop width
12 / Status · editorial
Each component gets 90 square cells, one per day, which is the most literal use of the grid anywhere in the system. Every cell has its own tooltip, every state has both a colour and a word, and incidents are listed newest first.
Get incident updates
Components · last 90 days
Edge network
OperationalBuilds
Operationalap-south-1
DegradedDashboard & API
OperationalIncident history
Elevated latency in ap-south-1
An upstream transit provider is dropping packets. Traffic has been shifted to secondary links; p95 is recovering.
Slow builds for large monorepos
A cache eviction bug forced cold builds. Fixed and backfilled.
Build queue outage
A failed database migration stopped the queue. Builds were retried automatically once it recovered. Read the postmortem →
Scroll sideways to see the whole preview at desktop width
This permanently removes the project, its services and 90 days of build history.