Keep watch from your own shore.

Self-hosted telemetry for Node apps. Adonis and NestJS in one command, any OpenTelemetry stack in one line.

Get started bunx vigie --seed
0 of 6 spottedTurn the light with your pointer.

From the first log line to the fix.

Logs, traces, task runs, issues and business processes, read on one clock.

Livelast 15 min

POST /parcels

5021.84 s, 7 spans
Open request
Span
Time
parcelo.ship
status
502 Bad Gateway
error.code
PARCELO_UNAVAILABLE
http.url
/parcelo/parcels
retry
scheduled in 30 s
Where the time went
external
1.42 s, 77%
database
118 ms, 3 queries
cache
8 ms
app
294 ms

parcelo.ship

succeededorder:7781
Open traceRequeue

parcelo.ship succeeded on attempt 2

Parcelo answered 502 the first time. The retry 30 s later went through.

Attempt 1 failedPARCELO_UNAVAILABLE, 1.42 s
waited 30 s
Attempt 2 succeeded201 Created, 0.61 s
Payload
{
  "orderId": 7781,
  "customer": "c***@acme.fr",
  "token": "[REDACTED]",
  "sessionCode": "RS6731-B2"
}
Part of
process
Order fulfilment
step
Shipment created at Parcelo
caused by
paynest.payment.ingest
queue wait
2.1 s

Issues

3 open
Parcelo is refusing callsPARCELO_UNAVAILABLE, 38 events
The request did not pass validationE_VALIDATION_ERROR, 6 events
Same query 14 times on GET /sessions/:idLikely N+1, 211 requests
firingfirst seen 13:21, last 14:02

Parcelo is refusing calls

Why

Parcelo did not accept the call: 502 Bad Gateway

Fix

Nothing yet. Tasks retry on their own. If it lasts, check the provider's status and lower this kind's concurrency.

ResolveCreate Linear issueAsk Vigie

Order fulfilment

From a Paynest payment to a parcel on its way. SLA 4 h.
today
Payment confirmed by Paynest412
Report rendered409
Customer emailed405
Shipment created at Parcelo371
6 waiting past 15 min, all on PARCELO_UNAVAILABLE
Declared to Taxbox398
Waiting at Shipment created at Parcelo
order:7781attempt 2 in 12 s22 min
order:7779attempt 3 in 1 min21 min
order:7764attempt 2 in 30 s19 min
order:7760attempt 2 in 44 s17 min
ClickHouse
Ingest
App
Alerts and MCP
definitions/

One Bun process. ClickHouse behind it.

Nothing else to operate. A broken definition shows its error and the rest keeps working.

30daysspans and metrics
90dayswide events
400daysrollups

Your app's words, in plain code.

Everything specific to your app lives in definitions/ and deploys with it. Click a highlighted value to change it.

definitions/renderers/providers.tssaved, reloaded
export const parcelo = defineLogRenderer({  id: "parcelo",  match: "parcelo.*",  title: "Parcelo",  render: (e) => [    ui.numbers({      observed: e.num("parcelo.parcelsObserved"),      created: e.num("parcelo.parcelsDelivered"),    }),    ui.fields(e, "parcelo.", {      tones: { refused:  },    }),  ],  row: (e) =>    `${e.num("parcelo.parcelsObserved")} parcels`,});

The event panel, for any event matching parcelo.*

Parcelo

parcelo.ship
42observed
41created
operation
deliver
outcome
refused
session
RS6731-B2
attempt
1

The log row, in place of raw JSON

14:02:11.4warnParcelo 42 parcels
definitions/renderers/platform.tssaved, reloaded
// A provider that stops answering is what most// task alerts come down to, so say that first.export const providerDown = defineErrorRenderer({  id: "provider-unavailable",  match: (error) => error.code.endsWith("_UNAVAILABLE"),  headline: (error) => `${name(error)} is refusing calls`,  recovered: (error) => `${name(error)} is answering again`,  why: (error) => `${name(error)} did not accept the call`,  fix: () => "Nothing yet. Tasks retry on their own.",});

The post in your channel

Vigie
14:02, 3 alerts grouped
Parcelo is refusing calls

Why

Parcelo did not accept the call: 502 Bad Gateway

What to do

Nothing yet. Tasks retry on their own.

Open issuePause for 1 h

Alerts within 30 s go out as one post, with the errors behind them.

definitions/processes/acme.tssaved, reloaded
export const fulfilment = defineProcess({  id: "results-delivery",  title: "Order fulfilment",  subject: "order",  sla: "4h",  steps: [    step.task("paynest.payment.ingest", { expectWithin: "30m" }),    step.parallel([      step.task("receipts.email", { optional: true }),      step.task("parcelo.ship", {        label: "Shipment created at Parcelo",        expectWithin: ,      }),    ]),  ],});

What is stuck, and why

Payment confirmed by Paynest412
Customer emailed405
Shipment created at Parcelo371
6 waiting past 15 min

Every step also gets a funnel, its instances, and an MCP tool that reads them.

definitions/alerts/acme.tssaved, reloaded
export const apiErrors = defineAlert({  id: "api-5xx",  name: "API 5xx rate",  when: { source: "request-error-rate" },  above: ,  window: "10m",  unit: "%",  condition: "Server errors above 2% over 10 min",});

The last two hours, against your threshold

Agents read the same screens you do.

POST /mcp speaks MCP over HTTP. Every page and every defineTool becomes a tool. Payloads never leave through it.

With a read token

  • get_context_pack
  • search_logs
  • get_event
  • get_trace
  • list_issues
  • list_tasks
  • list_processes
  • list_alerts
  • run_query

With a write token, as the signed-in person

  • resolve_issue
  • pause_alert

Vigie MCP
POST /mcp with a read token
11 tools, every page included
payloads stay on your server

Why are results not reaching Parcelo?
get_context_pack()Acme: 5 processes, 41 task kinds, 7 alerts
list_issues({ status: "open" })Parcelo is refusing calls, 38 events since 13:21
list_processes({ id: "order-fulfilment" })6 waiting at Shipment created at Parcelo
Parcelo has answered 502 since 13:21. Six deliveries are waiting and will retry on their own. Nothing to requeue yet; if it lasts past an hour, lower parcelo.ship's concurrency.

Secrets stop at ingest.

Redaction runs before anything is stored, so nothing downstream ever sees the raw value. Type one in.

What your app sendsWhat Vigie stores
What your app sendseditable
What Vigie stores
0secret field names
0paths with a secret
0emails in values
0card numbers
  • ClickHouse has no published port. Vigie runs as a non-root user on a read-only filesystem.
  • Keys are hashed at rest. Ingest keys, API tokens and MCP tokens are shown once.
  • Sign-up closes after the first owner, who needs a one-time code from the logs.
  • Raw SQL is refused if Vigie's database role can read credentials.

Running before your coffee cools.

One command starts Vigie. ClickHouse comes from an installed binary, Docker, or a download, and ports that are taken move to the next free one.

  1. Run Vigie. On your machine with no sign-in, or on a server with HTTPS.
  2. Connect your app. The installer finds Vigie, writes the env, and links the app in the browser, so nobody copies a key.
  3. Open a request. Every request, error and job is one event, joined to its trace.
bunx vigie --seed

# the app on http://localhost:5310, ingest on :4318 # --seed sends a day of realistic telemetry # no sign-in, listens on this machine only

deploy/install.sh --domain vigie.acme.dev

# Docker, ClickHouse and HTTPS from Caddy # prints the owner's setup code and the first ingest key

node ace add @vigiejs/adonis

# traces from @adonisjs/otel, one wide event per request # background work through task()

npx @vigiejs/nestjs init

# adds VigieModule.forRoot() and the trace register # @VigieTask() on BullMQ processors and crons

OTEL_EXPORTER_OTLP_ENDPOINT=http://localhost:4318

# any OpenTelemetry SDK, any language # a log with method, status and durationMs is a request