Status

Three things we can honestly report on, and one we won't.

Undercut is a static site plus one serverless function today: there's no server fleet, no SLA, and nothing that would make an “uptime %” figure honest. Below is Build — the real pass/fail history of our CI, live from GitHub — plus two sections, Routing intelligence and Service, that will report live data once Pro's routing backend ships. They're shown now, clearly marked pending, so this page's shape doesn't change out from under anyone later.

BuildLive · GitHub Actions

Checking latest run…

Live status is unavailable right now — GitHub's API didn't respond (likely a public rate limit; this page makes an unauthenticated request).

See the run history directly on GitHub instead:

Last 20 runs (newest first)

Full history on GitHub Actions ↗ Back to Undercut

Each dot is one run of .github/workflows/evals.yml (unit tests for the routing policy, plus a mocked eval-harness smoke run). Green = passed, red = failed, amber = another result (cancelled, in progress, etc). Data comes from api.github.com/repos/undercutsh/firstpass/actions/workflows/evals.yml/runs at page load — it is not cached or edited by hand.

Routing intelligenceComing with Pro

Policy artifact versionThe policy_version currently being served on the pro and teams:* channels. Pending backend
Signature validityWhether the live artifact's ed25519 signature verifies clean against the published key. Pending backend
Roster coverageShare of a Teams org's connected roster the latest artifact was compiled for. Pending backend

There is no Pro routing backend live yet, so there is nothing here to check — these three rows are placeholders, not a stand-in for real numbers. Once Pro ships, this section will read live: see the policy file for what a verified artifact and its signature actually mean.

ServiceComing with Pro

Basic reachabilityWhether the policy-fetch endpoint Pro/Teams clients check for a fresh artifact is answering requests. Pending backend

Same caveat as above: there's no endpoint running today to be reachable or not. This will become a real, live reachability check once one exists.

What we don't publish, and why

No uptime percentage appears anywhere on this page, and won't once Pro ships either. An honest uptime number requires a request path the number is actually measuring — and by design, Undercut mostly isn't in that path. A client that can't reach us falls back to its last verified cached policy, and if it has none, to the free static models.md tier→model map shipped in the skill itself, offline, no network required (see “What happens on a failed or expired verification” in the policy file). Since routing keeps working — just on slightly staler information — a percentage here would describe our servers, not whether coding agents kept dispatching correctly. We'd rather show three honest pending rows than one dishonest healthy one.