TAG platform status

Every service on preview.tag.company, tier by tier, with what was checked, what it answered, and how it has behaved since.

Running, with something degradedchecked 39s ago · 2026-11-21 13:32:32Z← Front door

This page is /status, and it is the only page on this estate that says whether anything is working. / is the way IN — names, purposes and links, and not one claim about health, because a fact with two owners is how this estate keeps being wrong. Machines poll /health, which answers this page as JSON with a status code that means something. The dot on a card at / is this page's verdict rendered again — the same snapshot, the same function — so it is the only claim that page makes and it cannot be a second opinion.

sampledevery 60s, in the background
this observation39s old
history since56d ago
samples stored199,732
survived17 restart(s)
degraded · 3ok · 43not deployed · 11held · 2internal only · 1not checked · 2
Needs attention. These are ours:
  • TrueForge chat — HTTP 200, expected 401
External dependency failing. Somebody else's, not this platform's, so it does not make /health red — it is answering 503. It is still what breaks first:
  • TAG login (tag-aio) — HTTP 307, expected 200/302
  • Hydra (tag-aio) — HTTP 404, expected 200

Web surfaces

degraded · 1ok · 7not deployed · 4held · 2internal only · 1

What a person can open in a browser. Probed THROUGH THE EDGE, at their public paths, so a green card means the routing works and not merely that a container is alive. A route that 404s is the failure this tier exists to catch — it is how /health itself was found missing. One surface here is NOT on this origin and says so on its own row; it is grouped by what a person does, not by where it is served from.

Origin (the TAG app)ok
/aletia
aletia

Under a prefix, which is why the front door can have /: two things cannot own one path. The prefix is compiled into the bundle by Next.js basePath, so it is a rebuild of aletia and not a routing change. See ../docs/nextjs-basepath.md. aletia replaced tag-mvp-ux as this image's source; each environment runs tag-mvp-ux's last image until its pin is moved to aletia's, in place (the Aletia row).

14:00Z24 of 24 hours observednow
state
ok for 3d · since 23:40Z
24 hours
100% 1,413 samples
7 days
99.97% 10,056 samples · 2 down, 1 degraded
30 days
99.99% 43,178 samples · 3 down, 2 degraded
all time
99.99% 43,178 samples · 3 down, 2 degraded · observed 56d
response
146 ms now p50 149 ms p95 235 ms · 1,440 samples, 24h
last failure
3d ago · down — The read operation timed out
shipped by
aletia, on its own merge — never from this repository
probe
GET /aletia · reproduce: curl -i https://preview.tag.company/aletia
expects HTTP 200
through the edge
depends on
Edge (Traefik) and TLSKong (API gateway)Dynamic.xyz origin allowlist
HTTP 200 · through the edge · 146 ms
Aletianot deployed
aletia

Not deployed in this environment. aletia replaces tag-mvp-ux at cutover, at Origin's address; until then Origin is what answers there and this row is not probed.

state
not deployed
shipped by
aletia, on its own merge — never from this repository

Never probed, so there is nothing to remember. Its state is read from the catalogue.

TAG Harnessnot deployed
tag-harness

This environment runs no TAG Harness: /chat on Coolify preview is TrueForge's own UI and stays so until that environment changes. The tag-aio test VM runs one at its own host; its row is in the tag-aio tier below.

state
not deployed
shipped by
tag-harness publishes the image by digest from its main, and infra/ovh pins it and runs it on the apps server where harness_host is set
depends on
Edge (Traefik) and TLSHydra (OAuth)

Never probed, so there is nothing to remember. Its state is read from the catalogue.

TrueForge chatdegraded
/chat
truefoundry/trueforge

Third-party (truefoundry/trueforge, pinned in trueforge/VERSION). BEHIND HTTP BASIC AUTHENTICATION AT THE EDGE since 2026-10-14, because the service itself authenticates nobody: every request, with no credential or with an invalid one, is served as the same `admin` subject, and the session list was readable in full by anyone with the URL. One shared password is not per-user access and not per-client isolation — everyone who has it still sees every conversation in this instance. See ../docs/chat-basic-auth.md. /chat on Coolify preview stays TrueForge until that environment changes; the TAG Harness that replaces it is on the tag-aio test VM, reported in the tag-aio tier below.

14:00Z24 of 24 hours observednow
state
degraded for 37d · since 00:02Z
24 hours
0.0% 1,413 samples · 1,413 degraded
7 days
0.0% 10,056 samples · 10,056 degraded
30 days
0.0% 43,178 samples · 43,178 degraded
all time
0.0% 43,178 samples · 43,178 degraded · observed 56d
response
259 ms now p50 107 ms p95 192 ms · 1,440 samples, 24h
last failure
0s ago · degraded — HTTP 200, expected 401
shipped by
this repository — the pin in trueforge/VERSION, the TAG theme and the mark, digested into one image tag
probe
GET /chat/healthz · reproduce: curl -i https://preview.tag.company/chat/healthz
expects HTTP 401
through the edge
depends on
Edge (Traefik) and TLSTrueForge PostgresTrueForge RedisOpenRouter (JEV_API_KEY)
HTTP 200, expected 401 · through the edge · 259 ms
TAG loginnot deployed
tag-platform/login

This environment runs no TAG login app: sign-in is Dynamic alone.

state
not deployed
shipped by
this repository: login/ publishes the image by digest (pull request #72), and infra/ovh pins it beside Kratos where the login host is set
depends on
Edge (Traefik) and TLSKratos (identity)Hydra (OAuth)

Never probed, so there is nothing to remember. Its state is read from the catalogue.

TAG Staff (Paperclip)not deployed
paperclipai/paperclip

Not reported here: STAFF_URL is not set in this health container, so this page does not know TAG Staff's host. Set it in the health service's environment to report the row.

14:00Z0 of 24 hours observednow
state
not deployed for 3d · since 23:51Z
uptime
nothing measurable yet · 279 sample(s), none of which measured up or down
last failure
none since this page started watching it
shipped by
this repository — the upstream digest in paperclip/DIGEST, the TAG theme and the mark, digested into one image tag
depends on
Postgres
Status pagelive
/status
tag-platform

The detail page — the one a person opens when a card here is not green. Probed at /health/live, which probes nothing itself, so the check cannot recurse into the whole estate every time somebody refreshes.

14:00Z24 of 24 hours observednow
state
live for 3d · since 08:11Z
24 hours
100% 1,413 samples
7 days
99.92% 10,056 samples · 8 degraded
30 days
99.97% 43,178 samples · 12 degraded
all time
99.97% 43,178 samples · 12 degraded · observed 56d
response
116 ms now p50 153 ms p95 253 ms · 1,440 samples, 24h
last failure
3d ago · degraded — HTTP 404, expected 200
running
this container does not report a commit
shipped by
this repository, on a merge to develop
probe
GET /status/live · reproduce: curl -i https://preview.tag.company/status/live
expects HTTP 200
through the edge
depends on
Edge (Traefik) and TLS
HTTP 200 — liveness only — not this service's own verdict · through the edge · 116 ms
Health endpointlive
/health
tag-platform

The same data as JSON, for a monitor rather than a person: 200 while everything that is SUPPOSED to be running is running, 503 when it is not. The entries below that are held or not deployed never make it red — an endpoint that is permanently 503 is an endpoint nobody reads.

14:00Z24 of 24 hours observednow
state
live for 3d · since 08:11Z
24 hours
100% 1,413 samples
7 days
99.92% 10,056 samples · 8 degraded
30 days
99.97% 43,178 samples · 12 degraded
all time
99.97% 43,178 samples · 12 degraded · observed 56d
response
173 ms now p50 153 ms p95 249 ms · 1,440 samples, 24h
last failure
3d ago · degraded — HTTP 404, expected 200
running
this container does not report a commit
shipped by
this repository, on a merge to develop
probe
GET /health/live · reproduce: curl -i https://preview.tag.company/health/live
expects HTTP 200
through the edge
depends on
Edge (Traefik) and TLS
HTTP 200 — liveness only — not this service's own verdict · through the edge · 173 ms
Agent fleet dashboardok
/agents
aletia/agents/tag-ui

tag-ui, deployed 2026-09-22. READ THE NEXT ROW WITH IT: what is published here is the dashboard and the four paths it serves itself, not the fleet. It calls tag-entry and tag-moderator from inside the Docker network, server-side; the agents' own endpoints (/agents/dispatch, /agents/match, /agents/fleet) answer 404 through this prefix.

14:00Z24 of 24 hours observednow
state
ok for 3d · since 23:40Z
24 hours
100% 1,413 samples
7 days
99.97% 10,056 samples · 3 degraded
30 days
99.99% 43,178 samples · 6 degraded
all time
99.99% 43,178 samples · 6 degraded · observed 56d
response
107 ms now p50 130 ms p95 222 ms · 1,440 samples, 24h
last failure
3d ago · degraded — HTTP 503, expected 200
shipped by
aletia/agents/tag-ui, on its own merge
probe
GET /agents/health · reproduce: curl -i https://preview.tag.company/agents/health
expects HTTP 200
through the edge
depends on
Edge (Traefik) and TLStag-entrytag-moderator
HTTP 200 · through the edge · 107 ms
Agent fleet APIinternal only
What /agents would expose if this were routed · aletia/agents

STILL INTERNAL ON PURPOSE, and this is the row that carries the security decision. The fleet authenticates no caller, so publishing its API would hand anyone on the internet the ability to drive it and spend its model budget. That a UI now sits at /agents does not change it, and must not be read as having changed it. Every agent is running and is reported, from inside the network, in the three Agent fleet tiers below. See ../docs/agents-exposure.md.

state
internal only
shipped by
aletia/agents, on its own merge
depends on
tag-entry

Never probed, so there is nothing to remember. Its state is read from the catalogue.

Truss client portalok
/client
truss-interfaces/apps/portal

SERVES ITS OWN PREFIX AND THE EDGE MUST NOT STRIP IT. `/client` is compiled into the image by Next.js basePath, so moving it is a rebuild in truss-interfaces and not a routing change here. Routed by a committed label in ../truss/compose.yaml; the file-provider router for the same prefix stays held, because two definitions of one route is undefined. See ../docs/truss-portals.md.

14:00Z24 of 24 hours observednow
state
ok for 30d · since 22:56Z
24 hours
100% 1,413 samples
7 days
100% 10,056 samples
30 days
100% 43,178 samples
all time
100% 43,178 samples · observed 56d
response
151 ms now p50 180 ms p95 273 ms · 1,440 samples, 24h
last failure
30d ago · degraded — HTTP 404, expected 200
shipped by
truss-interfaces, once it is built with a basePath
probe
GET /client · reproduce: curl -i https://preview.tag.company/client
expects HTTP 200
through the edge
HTTP 200 · through the edge · 151 ms
Truss firm portalok
/firm
truss-interfaces/apps/firm

A SEPARATE IMAGE from the client portal, not the same one served twice — each app compiles its own basePath, so the firm image answers 404 at /client and the portal image answers 404 at /firm. The same prefix rule as the row above, and the same reason its router is held.

14:00Z24 of 24 hours observednow
state
ok for 30d · since 22:56Z
24 hours
100% 1,413 samples
7 days
100% 10,056 samples
30 days
100% 43,178 samples
all time
100% 43,178 samples · observed 56d
response
138 ms now p50 170 ms p95 256 ms · 1,440 samples, 24h
last failure
30d ago · degraded — HTTP 404, expected 200
shipped by
truss-interfaces, once it is built with a basePath
probe
GET /firm · reproduce: curl -i https://preview.tag.company/firm
expects HTTP 200
through the edge
HTTP 200 · through the edge · 138 ms
Back-officeheld
/backoffice
tag-backoffice

Held by the captain's decision — "for tag-Gateway and /backoffice lets wait a little". Nothing technical is in the way.

state
held
shipped by
tag-backoffice, when the captain releases the hold

Never probed, so there is nothing to remember. Its state is read from the catalogue.

TAG-Gatewayheld
/gateway
tag-gateway

Held by the captain's decision, the same sentence. Reserved, not overlooked.

state
held
shipped by
tag-gateway, when the captain releases the hold

Never probed, so there is nothing to remember. Its state is read from the catalogue.

Marketing siteok
www.tag.company
tag-website

Served at the company's root domain by tag-website's own hosting, not by this edge. The front door keeps / on this host.

14:00Z24 of 24 hours observednow
state
ok for 1d · since 15:45Z
24 hours
100% 1,413 samples
7 days
not observed that long · 99.92% of 4,865 so far
30 days
not observed that long · 99.92% of 4,865 so far
all time
99.92% 4,865 samples · 4 down · observed 3d
response
317 ms now p50 355 ms p95 517 ms · 1,440 samples, 24h
last failure
1d ago · down — The read operation timed out
shipped by
tag-website, on its own merge, to its own hosting
probe
http · the address is not published on this page
expects HTTP 200
outbound
HTTP 200 · outbound · 317 ms

Data tier — Supabase

ok · 14

Self-hosted Supabase, probed component by component rather than as one "Supabase" row: the failure worth showing is the day PostgREST is fine and Realtime is not, and a single row hides exactly that. Kong is checked through the edge because it is the only published door; everything behind it is checked from inside, where it actually lives.

Kong (API gateway)ok
/supabase
The only published door to the tier — /supabase
14:00Z24 of 24 hours observednow
state
ok for 56d · since 00:19Z
24 hours
100% 1,413 samples
7 days
100% 10,056 samples
30 days
100% 43,178 samples
all time
100% 43,178 samples · observed 56d
response
79 ms now p50 142 ms p95 241 ms · 1,440 samples, 24h
last failure
none since this page started watching it
shipped by
Coolify's Supabase stack — pasted compose, not a repository
probe
GET /supabase/rest/v1/ · reproduce: curl -i https://preview.tag.company/supabase/rest/v1/
expects HTTP 200 or 401
through the edge
depends on
Edge (Traefik) and TLS
HTTP 401 · through the edge · 79 ms
PostgRESTok
/supabase/rest/v1
14:00Z24 of 24 hours observednow
state
ok for 56d · since 00:19Z
24 hours
100% 1,413 samples
7 days
100% 10,056 samples
30 days
100% 43,178 samples
all time
100% 43,178 samples · observed 56d
response
296 ms now p50 313 ms p95 395 ms · 1,440 samples, 24h
last failure
none since this page started watching it
probe
http · the address is not published on this page
expects HTTP 200 or 401 or 404
inside the network
depends on
Postgres
HTTP 200 · inside the network · 296 ms
GoTrue (auth)ok
/supabase/auth/v1
14:00Z24 of 24 hours observednow
state
ok for 56d · since 00:19Z
24 hours
100% 1,413 samples
7 days
100% 10,056 samples
30 days
100% 43,178 samples
all time
100% 43,178 samples · observed 56d
response
50 ms now p50 70 ms p95 163 ms · 1,440 samples, 24h
last failure
none since this page started watching it
probe
http · the address is not published on this page
expects HTTP 200
inside the network
depends on
Postgres
HTTP 200 · inside the network · 50 ms
Realtimeok
/supabase/realtime/v1 — WebSocket

403 here is HEALTHY: Realtime answered the upgrade and refused it because the page sends no apikey, which is what it should do. What this probe exists to catch is the opposite — a plain GET returns something perfectly plausible while subscriptions are broken.

14:00Z24 of 24 hours observednow
state
ok for 56d · since 00:19Z
24 hours
100% 1,413 samples
7 days
100% 10,056 samples
30 days
100% 43,178 samples
all time
100% 43,178 samples · observed 56d
response
65 ms now p50 15 ms p95 47 ms · 1,440 samples, 24h
last failure
none since this page started watching it
probe
websocket · the address is not published on this page
expects an upgrade handshake answering HTTP 101 or 403
inside the network
depends on
Postgres
HTTP 403 — serving, upgrade refused unauthenticated · inside the network · 65 ms
Storageok
/supabase/storage/v1
14:00Z24 of 24 hours observednow
state
ok for 56d · since 00:19Z
24 hours
100% 1,413 samples
7 days
100% 10,056 samples
30 days
100% 43,178 samples
all time
100% 43,178 samples · observed 56d
response
48 ms now p50 78 ms p95 171 ms · 1,440 samples, 24h
last failure
none since this page started watching it
probe
http · the address is not published on this page
expects HTTP 200
inside the network
depends on
PostgresStorage backend (MinIO)
HTTP 200 · inside the network · 48 ms
Postgresok
Not published. Reachable inside the network only.
14:00Z24 of 24 hours observednow
state
ok for 56d · since 00:19Z
24 hours
100% 1,413 samples
7 days
100% 10,056 samples
30 days
100% 43,178 samples
all time
100% 43,178 samples · observed 56d
response
13 ms now p50 7 ms p95 31 ms · 1,440 samples, 24h
last failure
none since this page started watching it
probe
tcp · the address is not published on this page
expects a TCP connection
inside the network
connected to :5432 · inside the network · 13 ms
Supavisor (pooler)ok
Connection pooler in front of Postgres
14:00Z24 of 24 hours observednow
state
ok for 56d · since 00:19Z
24 hours
100% 1,413 samples
7 days
100% 10,056 samples
30 days
100% 43,178 samples
all time
100% 43,178 samples · observed 56d
response
42 ms now p50 93 ms p95 188 ms · 1,440 samples, 24h
last failure
none since this page started watching it
probe
http · the address is not published on this page
expects HTTP 200 or 204
inside the network
depends on
Postgres
HTTP 204 · inside the network · 42 ms
Storage backend (MinIO)ok
Object store behind the Storage API
14:00Z24 of 24 hours observednow
state
ok for 56d · since 00:19Z
24 hours
100% 1,413 samples
7 days
100% 10,056 samples
30 days
100% 43,178 samples
all time
100% 43,178 samples · observed 56d
response
41 ms now p50 92 ms p95 183 ms · 1,440 samples, 24h
last failure
none since this page started watching it
probe
http · the address is not published on this page
expects HTTP 200
inside the network
HTTP 200 · inside the network · 41 ms
postgres-metaok
Schema introspection for Studio
14:00Z24 of 24 hours observednow
state
ok for 56d · since 00:19Z
24 hours
100% 1,413 samples
7 days
100% 10,056 samples
30 days
100% 43,178 samples
all time
100% 43,178 samples · observed 56d
response
58 ms now p50 95 ms p95 188 ms · 1,440 samples, 24h
last failure
none since this page started watching it
probe
http · the address is not published on this page
expects HTTP 200
inside the network
depends on
Postgres
HTTP 200 · inside the network · 58 ms
Studiook
Supabase's own dashboard. Deliberately not published.
14:00Z24 of 24 hours observednow
state
ok for 56d · since 00:19Z
24 hours
100% 1,413 samples
7 days
100% 10,056 samples
30 days
100% 43,178 samples
all time
100% 43,178 samples · observed 56d
response
106 ms now p50 106 ms p95 197 ms · 1,440 samples, 24h
last failure
none since this page started watching it
probe
http · the address is not published on this page
expects HTTP 200
inside the network
depends on
postgres-metaKong (API gateway)
HTTP 200 · inside the network · 106 ms
Edge Functionsok
Deno runtime. No health route — checked as a listening port.

supabase/edge-runtime answers 500 on /_internal/health and 400 on /, so neither is a health signal. The honest check is that it is listening; say that rather than dressing a 500 up as green.

14:00Z24 of 24 hours observednow
state
ok for 56d · since 00:19Z
24 hours
100% 1,413 samples
7 days
100% 10,056 samples
30 days
100% 43,178 samples
all time
100% 43,178 samples · observed 56d
response
15 ms now p50 6 ms p95 35 ms · 1,440 samples, 24h
last failure
none since this page started watching it
probe
tcp · the address is not published on this page
expects a TCP connection
inside the network
depends on
Postgres
connected to :9000 · inside the network · 15 ms
Analytics (Logflare)ok
Log sink the whole tier writes to
14:00Z24 of 24 hours observednow
state
ok for 56d · since 00:19Z
24 hours
100% 1,413 samples
7 days
100% 10,056 samples
30 days
100% 43,178 samples
all time
100% 43,178 samples · observed 56d
response
120 ms now p50 157 ms p95 240 ms · 1,440 samples, 24h
last failure
none since this page started watching it
probe
http · the address is not published on this page
expects HTTP 200
inside the network
depends on
Postgres
HTTP 200 · inside the network · 120 ms
Vectorok
Ships container logs into Logflare
14:00Z24 of 24 hours observednow
state
ok for 56d · since 00:19Z
24 hours
100% 1,413 samples
7 days
100% 10,056 samples
30 days
100% 43,178 samples
all time
100% 43,178 samples · observed 56d
response
80 ms now p50 103 ms p95 187 ms · 1,440 samples, 24h
last failure
none since this page started watching it
probe
http · the address is not published on this page
expects HTTP 200
inside the network
depends on
Analytics (Logflare)
HTTP 200 · inside the network · 80 ms
imgproxyok
Image transformation for Storage
14:00Z24 of 24 hours observednow
state
ok for 56d · since 00:19Z
24 hours
100% 1,413 samples
7 days
100% 10,056 samples
30 days
100% 43,178 samples
all time
100% 43,178 samples · observed 56d
response
118 ms now p50 107 ms p95 191 ms · 1,440 samples, 24h
last failure
none since this page started watching it
probe
http · the address is not published on this page
expects HTTP 200
inside the network
depends on
Storage backend (MinIO)
HTTP 200 · inside the network · 118 ms

Agent fleet — the way in

ok · 2

Everything reaching the fleet arrives through tag-entry. If this is not green, nothing below it will be doing anything, however green it looks — so read this tier first and the other two only if it is fine.

tag-entryok
Fleet entry point · :8200
14:00Z24 of 24 hours observednow
state
ok for 3d · since 23:30Z
24 hours
100% 1,413 samples
7 days
99.98% 10,056 samples · 2 down
30 days
99.99% 43,178 samples · 4 down
all time
99.99% 43,178 samples · 4 down · observed 56d
response
272 ms now p50 301 ms p95 384 ms · 1,440 samples, 24h
last failure
3d ago · down — [Errno 111] Connection refused
shipped by
aletia/agents, on its own merge
probe
http · the address is not published on this page
expects HTTP 200
inside the network
HTTP 200 · inside the network · 272 ms
tag-plannerok
Plans the work before it is handed out · :8102
14:00Z24 of 24 hours observednow
state
ok for 3d · since 23:30Z
24 hours
100% 1,413 samples
7 days
99.98% 10,056 samples · 2 down
30 days
99.99% 43,178 samples · 5 down
all time
99.99% 43,178 samples · 5 down · observed 56d
response
88 ms now p50 105 ms p95 183 ms · 1,440 samples, 24h
last failure
3d ago · down — [Errno -3] Temporary failure in name resolution
shipped by
aletia/agents, on its own merge
probe
http · the address is not published on this page
expects HTTP 200
inside the network
depends on
tag-entry
HTTP 200 · inside the network · 88 ms

Agent fleet — model-backed

ok · 3

THE THREE THAT CALL A MODEL, and the three that cost money to run. They share one credential with TrueForge — the estate's JEV_API_KEY, which is an OpenRouter key — so when classification and matching go quiet, the row to read next is OpenRouter in External dependencies, not these. That they answer at all is itself a signal: require_decision_config() runs in each one's startup hook and the process exits without the key, so a missing credential shows up here as down, not as silence.

tag-moderatorok
Decides where a mark goes · :8101 · calls a model
14:00Z24 of 24 hours observednow
state
ok for 3d · since 23:40Z
24 hours
100% 1,413 samples
7 days
99.97% 10,056 samples · 3 down
30 days
99.98% 43,178 samples · 7 down
all time
99.98% 43,178 samples · 7 down · observed 56d
response
101 ms now p50 104 ms p95 187 ms · 1,440 samples, 24h
last failure
3d ago · down — [Errno 111] Connection refused
shipped by
aletia/agents, on its own merge
probe
http · the address is not published on this page
expects HTTP 200
inside the network
depends on
tag-entryOpenRouter (JEV_API_KEY)
HTTP 200 · inside the network · 101 ms
tag-matcherok
Matches marks to expectations · :8103 · calls a model
14:00Z24 of 24 hours observednow
state
ok for 3d · since 23:40Z
24 hours
100% 1,413 samples
7 days
99.97% 10,056 samples · 3 down
30 days
99.98% 43,178 samples · 7 down
all time
99.98% 43,178 samples · 7 down · observed 56d
response
86 ms now p50 103 ms p95 182 ms · 1,440 samples, 24h
last failure
3d ago · down — [Errno 111] Connection refused
shipped by
aletia/agents, on its own merge
probe
http · the address is not published on this page
expects HTTP 200
inside the network
depends on
tag-entryOpenRouter (JEV_API_KEY)
HTTP 200 · inside the network · 86 ms
tag-arc-creatorok
Turns plans into arcs · :8107 · calls a model
14:00Z24 of 24 hours observednow
state
ok for 3d · since 23:40Z
24 hours
100% 1,413 samples
7 days
99.97% 10,056 samples · 3 down
30 days
99.98% 43,178 samples · 7 down
all time
99.98% 43,178 samples · 7 down · observed 56d
response
180 ms now p50 103 ms p95 177 ms · 1,440 samples, 24h
last failure
3d ago · down — [Errno 111] Connection refused
shipped by
aletia/agents, on its own merge
probe
http · the address is not published on this page
expects HTTP 200
inside the network
depends on
tag-entryOpenRouter (JEV_API_KEY)
HTTP 200 · inside the network · 180 ms

Agent fleet — document pipeline

ok · 4not deployed · 1

Deterministic processing: a document arrives, and these four carry it from bytes to a mark with an expectation attached. They call no model, so an OpenRouter outage does not touch them — if documents stop moving while the tier above is green, the fault is here.

tag-ingestionok
Takes the document in · :8104
14:00Z24 of 24 hours observednow
state
ok for 3d · since 23:40Z
24 hours
100% 1,413 samples
7 days
99.97% 10,056 samples · 3 down
30 days
99.98% 43,178 samples · 7 down
all time
99.98% 43,178 samples · 7 down · observed 56d
response
136 ms now p50 103 ms p95 179 ms · 1,440 samples, 24h
last failure
3d ago · down — [Errno 111] Connection refused
shipped by
aletia/agents, on its own merge
probe
http · the address is not published on this page
expects HTTP 200
inside the network
depends on
tag-entry
HTTP 200 · inside the network · 136 ms
tag-extractionok
Pulls content out of it · :8105
14:00Z24 of 24 hours observednow
state
ok for 3d · since 23:40Z
24 hours
100% 1,413 samples
7 days
99.97% 10,056 samples · 3 down
30 days
99.98% 43,178 samples · 7 down
all time
99.98% 43,178 samples · 7 down · observed 56d
response
81 ms now p50 102 ms p95 172 ms · 1,440 samples, 24h
last failure
3d ago · down — [Errno 111] Connection refused
shipped by
aletia/agents, on its own merge
probe
http · the address is not published on this page
expects HTTP 200
inside the network
depends on
tag-entry
HTTP 200 · inside the network · 81 ms
tag-enrichmentok
Adds what the content implies · :8106
14:00Z24 of 24 hours observednow
state
ok for 3d · since 23:40Z
24 hours
100% 1,413 samples
7 days
99.97% 10,056 samples · 3 down
30 days
99.99% 43,178 samples · 6 down
all time
99.99% 43,178 samples · 6 down · observed 56d
response
67 ms now p50 100 ms p95 167 ms · 1,440 samples, 24h
last failure
3d ago · down — [Errno 111] Connection refused
shipped by
aletia/agents, on its own merge
probe
http · the address is not published on this page
expects HTTP 200
inside the network
depends on
tag-entry
HTTP 200 · inside the network · 67 ms
tag-expectationok
Attaches the obligation it satisfies · :8108
14:00Z24 of 24 hours observednow
state
ok for 3d · since 23:40Z
24 hours
100% 1,413 samples
7 days
99.97% 10,056 samples · 3 down
30 days
99.99% 43,178 samples · 6 down
all time
99.99% 43,178 samples · 6 down · observed 56d
response
154 ms now p50 102 ms p95 166 ms · 1,440 samples, 24h
last failure
3d ago · down — [Errno 111] Connection refused
shipped by
aletia/agents, on its own merge
probe
http · the address is not published on this page
expects HTTP 200
inside the network
depends on
tag-entry
HTTP 200 · inside the network · 154 ms
tag-anchornot deployed
The tail of the pipeline — publishes to a public network

Absent in the Coolify environment — it refuses to anchor without a bootstrapped OpenETR signing identity in a volume. Nothing here is anchored, and nothing upstream of it is affected: the pipeline runs to completion and stops one step short of publishing.

state
not deployed

Never probed, so there is nothing to remember. Its state is read from the catalogue.

Supporting services

ok · 2not deployed · 6not checked · 2

Things no path points at, that a published service stops working without. They are listed because "the chat is broken" is usually one of these, and an operator should not have to already know they exist.

TrueForge Postgresok
Holds TrueForge's sessions AND its provider credentials

Dropping this volume loses the OpenRouter provider configuration — it lives here, not in the container environment and not in this repository.

14:00Z24 of 24 hours observednow
state
ok for 3d · since 04:10Z
24 hours
100% 1,413 samples
7 days
99.99% 10,056 samples · 1 down
30 days
100.00% 43,178 samples · 1 down
all time
100.00% 43,178 samples · 1 down · observed 56d
response
21 ms now p50 9 ms p95 55 ms · 1,440 samples, 24h
last failure
3d ago · down — [Errno -3] Temporary failure in name resolution
probe
tcp · the address is not published on this page
expects a TCP connection
inside the network
connected to :5432 · inside the network · 21 ms
TrueForge Redisok
Queues and schedule state for the chat
14:00Z24 of 24 hours observednow
state
ok for 3d · since 04:10Z
24 hours
100% 1,413 samples
7 days
99.99% 10,056 samples · 1 down
30 days
100.00% 43,178 samples · 1 down
all time
100.00% 43,178 samples · 1 down · observed 56d
response
26 ms now p50 9 ms p95 51 ms · 1,440 samples, 24h
last failure
3d ago · down — [Errno -3] Temporary failure in name resolution
probe
tcp · the address is not published on this page
expects a TCP connection
inside the network
connected to :6379 · inside the network · 26 ms
TAG Staff databasenot checked
A `paperclip` database inside supabase-db, with its own role

NOT a second Postgres. It is a database of its own inside the existing supabase-db service, owned by a `paperclip` role that can reach nothing else — so a Paperclip migration cannot touch the product's tables, and a dump of one is not a dump of the other. Nothing on this page has a credential for it, so this page will not claim to have checked it: the Postgres row in the data tier is what says the server is up. Paperclip runs on 15.8 — established by running its migrations against supabase/postgres:15.8.1.085, not assumed from its compose, which pins 17. See ../paperclip/README.md.

state
not checked
shipped by
nothing — it is a database inside the Supabase service, created once by hand; paperclip/README.md has the three statements

Never probed, so there is nothing to remember. Its state is read from the catalogue.

TrueForge controllernot checked
Runs the schedule-dispatch loop, out of process

A worker with no listening port and no health endpoint. Nothing on the network can observe it, so this page will not pretend to: if schedules stop firing, read its container log. Reporting a guess as green is the failure this page is built to avoid.

state
not checked
shipped by
this repository, pinned to the same image as /chat
depends on
TrueForge PostgresTrueForge Redis

Never probed, so there is nothing to remember. Its state is read from the catalogue.

Xero MCP servernot deployed
Xero sign-in and the books, for TRUSS, the gateway, both portals and /chat

This environment uses the shared server in Azure, which this page does not probe: it is not on this edge.

state
not deployed
shipped by
xero-mcp-server's Build & Publish workflow publishes the image from its main; infra/ovh pins it and runs it on the apps server

Never probed, so there is nothing to remember. Its state is read from the catalogue.

Hydra (OAuth)not deployed
Issues the OAuth tokens the MCP clients, the Harness and TrueForge present. It holds clients, consents and keys, and no people

This environment runs no OAuth server: sign-in is Dynamic alone.

state
not deployed
shipped by
this repository: hydra/ publishes the image, and infra/ovh pins it and runs it on the data server where auth_host is set

Never probed, so there is nothing to remember. Its state is read from the catalogue.

Kratos (identity)not deployed
Holds people's identities, their passkeys and security keys, and their sessions, behind self-service registration, login, verification and recovery. No sign-in is wired to it yet

This environment runs no identity server: sign-in is Dynamic alone.

state
not deployed
shipped by
this repository: kratos/ publishes the image, and infra/ovh pins it and runs it on the data server where kratos_host is set

Never probed, so there is nothing to remember. Its state is read from the catalogue.

TrueForge (per organization)not deployed
One instance per organization, reached only by the Harness

This environment runs no TrueForge behind a Harness. /chat's TrueForge is the web row of that name.

state
not deployed
shipped by
infra/ovh, with the Harness: one TrueForge, Postgres database and Redis per organization, on the TrueForge pin in trueforge/VERSION
depends on
TAG Harness

Never probed, so there is nothing to remember. Its state is read from the catalogue.

Aletia MCPnot deployed
The Aletia connector for AI clients, on the MCP host

Defined and not applied in this environment: /mcp is a route nobody has applied (docs/mcp-public-path.md), so there is no connector URL to probe. The handler itself answers under Origin's prefix.

state
not deployed
shipped by
the app behind Origin's address (aletia, which replaced tag-mvp-ux), routed by infra/ovh where mcp_host is set
depends on
Origin (the TAG app)

Never probed, so there is nothing to remember. Its state is read from the catalogue.

TAG Back Office MCPnot deployed
The TAG Back Office connector for AI clients, on the MCP host at /tag

This environment runs no TAG Back Office, so it has no MCP of its own.

state
not deployed
shipped by
tag-backoffice, on its own merge; routed by infra/ovh where mcp_host is set

Never probed, so there is nothing to remember. Its state is read from the catalogue.

Platform

ok · 2

The edge itself and the control plane that deploys onto it. If the edge row is not green you are not reading this page, so treat it as the certificate's expiry clock — the quiet failure that takes the whole origin down on a date nobody has in their calendar.

Edge (Traefik) and TLSok
One origin, one certificate, one thing to renew
14:00Z24 of 24 hours observednow
state
ok for 56d · since 00:19Z
24 hours
100% 1,413 samples
7 days
100% 10,056 samples
30 days
100% 43,178 samples
all time
100% 43,178 samples · observed 56d
response
113 ms now p50 126 ms p95 178 ms · 1,440 samples, 24h
last failure
none since this page started watching it
shipped by
Coolify's own proxy — not from this repository
probe
tls · the address is not published on this page
expects a certificate a browser accepts, more than 21 days from expiry
through the edge
depends on
Platform DNSLet's Encrypt
valid, 89 days left · through the edge · 113 ms
Coolifyok
Deploys and supervises everything above. Its own host.

Not on this origin, and nothing here depends on it at request time — it is down that redeploys and restarts stop being possible, not the site.

14:00Z24 of 24 hours observednow
state
ok for 1d · since 00:58Z
24 hours
100% 1,413 samples
7 days
99.99% 10,056 samples · 1 degraded
30 days
100.00% 43,178 samples · 1 degraded
all time
100.00% 43,178 samples · 1 degraded · observed 56d
response
135 ms now p50 157 ms p95 214 ms · 1,440 samples, 24h
last failure
1d ago · degraded — HTTP 502, expected 200
probe
http · the address is not published on this page
expects HTTP 200
outbound
HTTP 200 · outbound · 135 ms

tag-aio (preview.tag.company)

degraded · 2ok · 2

The tag-aio test VM, which *.preview.tag.company points to: the TAG Harness, the TAG login, Hydra and its own status page. Reported on every environment, because it is the same VM wherever this page runs. A test VM being down is never this environment's outage, so this tier is degrade-only: a failure here reads degraded and never turns /health 503.

TAG Harness (tag-aio)live
harness.preview.tag.company · tag-harness

The TAG Harness on the tag-aio test VM. Its root answering proves the host routes to a live process, not that anybody can sign in.

14:00Z24 of 24 hours observednow
state
live for 1d · since 15:22Z
24 hours
100% 1,413 samples
7 days
not observed that long · 99.88% of 4,865 so far
30 days
not observed that long · 99.88% of 4,865 so far
all time
99.88% 4,865 samples · 6 degraded · observed 3d
response
156 ms now p50 179 ms p95 240 ms · 1,440 samples, 24h
last failure
1d ago · degraded — The read operation timed out
probe
http · the address is not published on this page
expects HTTP 200 or 302
outbound
HTTP 200 — liveness only — not this service's own verdict · outbound · 156 ms
TAG login (tag-aio)degraded
login.preview.tag.company · tag-platform/login

The TAG login app on the tag-aio test VM. Its root answering proves the host routes to a live process, not a sign-in.

14:00Z24 of 24 hours observednow
state
degraded for 3d · since 04:28Z
24 hours
0.0% 1,413 samples · 1,413 degraded
7 days
not observed that long · 0.0% of 4,865 so far
30 days
not observed that long · 0.0% of 4,865 so far
all time
0.0% 4,865 samples · 4,865 degraded · observed 3d
response
737 ms now p50 760 ms p95 837 ms · 1,440 samples, 24h
last failure
0s ago · degraded — HTTP 307, expected 200/302
probe
http · the address is not published on this page
expects HTTP 200 or 302
outbound
HTTP 307, expected 200/302 · outbound · 737 ms
Hydra (tag-aio)degraded
auth.preview.tag.company · ory/hydra

Ory Hydra on the tag-aio test VM, asked its own readiness endpoint, which answers 200 only when Hydra can reach its database.

14:00Z24 of 24 hours observednow
state
degraded for 3d · since 04:28Z
24 hours
0.0% 1,413 samples · 1,413 degraded
7 days
not observed that long · 0.0% of 4,865 so far
30 days
not observed that long · 0.0% of 4,865 so far
all time
0.0% 4,865 samples · 4,865 degraded · observed 3d
response
132 ms now p50 147 ms p95 202 ms · 1,440 samples, 24h
last failure
0s ago · degraded — HTTP 404, expected 200
probe
http · the address is not published on this page
expects HTTP 200
outbound
HTTP 404, expected 200 · outbound · 132 ms
Status page (tag-aio)live
status.preview.tag.company · tag-platform

The tag-aio test VM's own status page. Its front door answering proves the page is served, not what it reports.

14:00Z24 of 24 hours observednow
state
live for 3d · since 12:08Z
24 hours
100% 1,413 samples
7 days
not observed that long · 99.96% of 4,865 so far
30 days
not observed that long · 99.96% of 4,865 so far
all time
99.96% 4,865 samples · 2 degraded · observed 3d
response
177 ms now p50 180 ms p95 237 ms · 1,440 samples, 24h
last failure
3d ago · degraded — The read operation timed out
probe
http · the address is not published on this page
expects HTTP 200
outbound
HTTP 200 — liveness only — not this service's own verdict · outbound · 177 ms

External dependencies

ok · 7

Not ours, and the reason this page is worth building. Of the three things that blocked this estate on 2026-09-21, TWO were external — Dynamic.xyz's origin allowlist and Docker Hub refusing an image pull — and neither would appear in a check that only looks at our containers. So neither is checked by reachability here: the allowlist is checked by asking Dynamic whether THIS origin is on it, and Docker Hub by taking an anonymous pull token and reading the rate limit back. Anything failing here is DEGRADED rather than down, is listed under `external` in /health, and NEVER changes that endpoint's status code: what we run and what we depend on are counted separately, because an endpoint answering 503 on somebody else's bad afternoon is one everybody learns to ignore. The allowlist row is the single exception, and says why. Each row also declares which answer is healthy — a registry's 401 and Anthropic's 401 are both correct — because "is it 200" and "is it well" are different questions.

Dynamic.xyz origin allowlistok
Authentication. Login fails silently if this origin is not listed.

This asks the real question. On 2026-09-21 every container was healthy and login was impossible because this environment's origin was not on Dynamic's list; a reachability check was green throughout. The probe sends this origin and requires Dynamic to echo it back in Access-Control-Allow-Origin — which it only does for an allowed origin. The environment id is public: it ships in the browser bundle.

14:00Z24 of 24 hours observednow
state
ok for 12h · since 01:08Z
24 hours
99.86% 1,413 samples · 2 down
7 days
99.98% 10,056 samples · 2 down
30 days
99.99% 43,178 samples · 6 down
all time
99.99% 43,178 samples · 6 down · observed 56d
response
169 ms now p50 208 ms p95 303 ms · 1,440 samples, 24h
changed
2 time(s) in 24h — a service that flaps is a different problem from one that is down
last failure
12h ago · down — The read operation timed out
probe
cors · the address is not published on this page
expects Access-Control-Allow-Origin naming this origin — reachability is not the question
outbound
this origin is allowlisted · outbound · 169 ms
Docker Hub pull quotaok
Image registry — rate-limits anonymous pulls by IP

Reachability was never the problem. Coolify's Supabase template referenced minio/mc with no tag, Docker Hub answered "pull access denied", and because one failed pull aborts the whole `docker compose pull`, NOTHING started. This probe takes an anonymous pull token and reads the quota back, so the row goes amber while there is still headroom rather than on the deploy that fails.

14:00Z24 of 24 hours observednow
state
ok for 14d · since 23:48Z
24 hours
100% 1,413 samples
7 days
100% 10,056 samples
30 days
100.00% 43,178 samples · 2 degraded
all time
100.00% 43,178 samples · 2 degraded · observed 56d
response
383 ms now p50 402 ms p95 477 ms · 1,440 samples, 24h
last failure
14d ago · degraded — The read operation timed out
probe
registry · the address is not published on this page
expects an anonymous pull token for ratelimitpreview/test and a quota that is not spent
outbound
83 pulls left of 100/hr · outbound · 383 ms
quay.iook
Image registry — where minio/mc actually comes from

401 ON THE MANIFEST IS THIS REGISTRY WORKING, and this row said "degraded" for it. quay.io answers /v2/ with a bearer challenge and issues an anonymous pull token — a complete, healthy handshake — and then refuses minio/mc, because MinIO's community repositories are no longer anonymously readable on quay.io OR on Docker Hub. That is MinIO's access policy, not quay.io's health, and until the expected statuses below were declared this page reported it as an outage of theirs and (through the verdict) as an outage of ours. Re-establish it rather than trusting this paragraph: curl -s https://quay.io/v2/ -o /dev/null -w '%{http_code}\n' — 401 with a WWW-Authenticate header is the healthy answer. The thing this row cannot tell you is whether a redeploy of the Supabase stack can still pull minio/mc; it cannot, anonymously, and the Docker Hub row above is where that first showed up.

14:00Z24 of 24 hours observednow
state
ok for 8d · since 20:23Z
24 hours
100% 1,413 samples
7 days
100% 10,056 samples
30 days
99.93% 43,178 samples · 29 degraded
all time
99.93% 43,178 samples · 29 degraded · observed 56d
response
394 ms now p50 428 ms p95 504 ms · 1,440 samples, 24h
last failure
8d ago · degraded — The read operation timed out
probe
registry · the address is not published on this page
expects an anonymous pull token for minio/mc and a quota that is not spent
outbound
registry serving; minio/mc is not anonymously readable (HTTP 401, as declared) · outbound · 394 ms
OpenRouter (JEV_API_KEY)ok
Model access for tag-matcher, tag-moderator, tag-arc-creator

The credential the estate carries as JEV_API_KEY is an OpenRouter key (sk-or-v1-…), and TrueForge's provider uses the same one. Named here for what it is, because a row labelled only "Jev API" sends whoever is debugging a 401 to the wrong dashboard. Reachability only — the model list, which spends no quota.

14:00Z24 of 24 hours observednow
state
ok for 9d · since 17:20Z
24 hours
100% 1,413 samples
7 days
100% 10,056 samples
30 days
100.00% 43,178 samples · 1 degraded
all time
100.00% 43,178 samples · 1 degraded · observed 56d
response
97 ms now p50 110 ms p95 168 ms · 1,440 samples, 24h
last failure
9d ago · degraded — The read operation timed out
probe
http · the address is not published on this page
expects HTTP 200
outbound
HTTP 200 · outbound · 97 ms
Anthropic APIok
LLM provider for the fleet (ANTHROPIC_API_KEY)
14:00Z24 of 24 hours observednow
state
ok for 8d · since 10:16Z
24 hours
100% 1,413 samples
7 days
100% 10,056 samples
30 days
99.99% 43,178 samples · 6 degraded
all time
99.99% 43,178 samples · 6 degraded · observed 56d
response
165 ms now p50 139 ms p95 194 ms · 1,440 samples, 24h
last failure
8d ago · degraded — HTTP 529, expected 401/403
probe
http · the address is not published on this page
expects HTTP 401 or 403
outbound
HTTP 401 · outbound · 165 ms
Let's Encryptok
Certificate issuance and renewal

Issuance, not the certificate in use — that is the Edge row in the Platform tier, which counts the days left on the one being served.

14:00Z24 of 24 hours observednow
state
ok for 9d · since 06:37Z
24 hours
100% 1,413 samples
7 days
100% 10,056 samples
30 days
100.00% 43,178 samples · 2 degraded
all time
100.00% 43,178 samples · 2 degraded · observed 56d
response
195 ms now p50 226 ms p95 291 ms · 1,440 samples, 24h
last failure
9d ago · degraded — _ssl.c:1001: The handshake operation timed out
probe
http · the address is not published on this page
expects HTTP 200
outbound
HTTP 200 · outbound · 195 ms
Platform DNSok
Does the platform host resolve?
14:00Z24 of 24 hours observednow
state
ok for 56d · since 00:19Z
24 hours
100% 1,413 samples
7 days
100% 10,056 samples
30 days
100% 43,178 samples
all time
100% 43,178 samples · observed 56d
response
65 ms now p50 7 ms p95 30 ms · 1,440 samples, 24h
last failure
none since this page started watching it
probe
dns · the address is not published on this page
expects a name that resolves
outbound
resolves · outbound · 65 ms