KeyNest Waitlist

Observability & cost

See each database's health, what it will cost this month, and why — without an external monitoring stack.

Every ready database streams usage telemetry into the KeyNest beta dashboard: time-series charts, percentile latency, a plain-language health verdict, and a projected month-end cost with the reasoning behind it. Native alerting and zero-config export are on the beta tier — no paid gate, no Grafana or Datadog to stand up. All cost figures are a labeled hypothesis, never a per-request bill or an SLA.

Usage telemetry Memory, connections, commands, and ingress/egress over time, with selectable 1h / 24h / 7d / 30d ranges and a freshness badge.
Percentile latency p50 / p95 / p99 shown numerically and over time, against a beta target line.
Projected cost A month-end projection on the Overview and per database, with a "why this cost" breakdown — flat plan rate, not per request.
Native alerting In-console threshold rules that fire to a dashboard feed and email, with zero external tooling.

Health verdict and freshness

  • healthy — memory and connections both under 75% of the plan ceiling.
  • watch — memory or connections at or above 75%.
  • saturated — memory or connections at or above 90%; consider resizing.
  • A freshness badge shows the last sample time; evidence older than 15 minutes is flagged stale, never shown as a silent gap.

Percentile latency methodology

  • Latency is observed client round-trip time sampled by the KeyNest collector — an honest signal, not a server-side SLA percentile.
  • The chart shows p50 / p95 / p99 against a beta target line (currently 2 ms p99). The target is a goal, not a guarantee.
  • Latency feeds the alerting and export surfaces, so you can alert on or export p99 without extra setup.

Cost transparency

  • Projected month-end cost is derived from your flat beta plan rate and the elapsed billing period — labeled hypothesis.
  • The "why this cost" breakdown ties the figure to the plan rate and your observed usage headroom.
  • Cost does not scale per request: more commands or connections within your plan do not change the number. That is the explicit contrast with request-metered providers.

Native threshold alerting

  • Rule types: memory %, connection %, p99 latency, command rate, projected cost, and evidence staleness.
  • Suggested one-click rules (memory > 80%, p99 > target, evidence stale > 30m) get you started.
  • Rules are evaluated server-side on each telemetry sample; a sustained breach fires once — to an in-dashboard feed and an email to your workspace contacts — until it resolves.
  • Alert payloads carry only metric, threshold, observed value, and database label. No credentials, connection strings, or provider diagnostics.

Zero-config usage & cost export

  • Create a per-client export token in the dashboard Usage section. It is shown once, stored only as a hash, and revocable — revoking it stops every scrape and download immediately.
  • Prometheus: a read-only scrape endpoint exposing per-database usage and projected cost gauges. No paid tier, no data-source dance.
  • CSV: a range export including usage columns and cost line items, with a hypothesis caveat on every row.
  • Exports contain no secrets or provider-internal fields.
Export endpoints (authenticate with your export token) 
# Prometheus scrape — paste the URL + token into your scrape config
GET https://api.keynest.dev/v0/phase3/export/prometheus?client_id=<client>
Authorization: Bearer <export-token>

# Gauges (labeled client + database):
#   keynest_db_used_memory_mb        keynest_db_connected_clients
#   keynest_db_commands_total        keynest_db_latency_p50_ms
#   keynest_db_latency_p99_ms        keynest_db_projected_cost_usd
#   keynest_db_sample_age_seconds    ...

# CSV — usage + cost line items for a range (1h | 24h | 7d | 30d)
GET https://api.keynest.dev/v0/phase3/export/usage.csv?client_id=<client>&range=24h
Authorization: Bearer <export-token>

Not in scope during beta

  • Per-request or usage-based billing, invoicing, or a GA price — cost stays a labeled hypothesis.
  • Full APM/tracing, raw-argument slow logs, or arbitrary in-console PromQL.
  • Push-based export / remote-write and Datadog/Grafana-specific integrations.
  • SLA, compliance, or durability commitments (unchanged from beta compatibility).

Open the live dashboard at /dashboard (accepted invite required), or preview the surfaces with sample data at /dashboard?demo=1. Compatibility boundaries: beta compatibility guide.