Pulsegrid / Platform

Platform

Three parts: an agent that runs on each host, a columnar store, and a query layer that speaks SQL. This page describes what each one does and what it does not.

The agent

A single static binary, about 34MB, no runtime dependencies. It runs as a systemd unit or a DaemonSet and needs read access to /proc and the container socket.

Runtime	Auto-instrumented	Method
Go 1.21+	HTTP, gRPC, database/sql	eBPF uprobes, no rebuild
Node 18+	HTTP, Express, Fastify, pg, Redis	Loader hook, no code change
Python 3.9+	Django, Flask, FastAPI, psycopg, requests	sitecustomize injection
Java 11+	Servlet, Spring, JDBC, Kafka	JVM agent, one flag
Ruby 3.0+	Rails, Sidekiq, ActiveRecord	Gem, one require
.NET 6+	ASP.NET Core, HttpClient, EF Core	CLR profiler
Anything else	Via OpenTelemetry OTLP	Point your exporter at the agent

Agent overhead measured on our own production fleet is 0.4% to 1.1% CPU and roughly 90MB resident, depending on trace volume.

Storage and retention

Signal	Resolution	Solo	Team	Scale
Metrics, raw	10 second	2 days	15 days	30 days
Metrics, rolled up	1 minute	7 days	90 days	13 months
Traces, sampled	Full fidelity	7 days	30 days	90 days
Traces, errored	Always kept	7 days	90 days	13 months
Logs	Full text	3 days	30 days	90 days
Deploy and event markers	Full fidelity	90 days	13 months	Indefinite

Retention is per signal, not a shared quota, and there is no ingest cap on any tier. Extended retention beyond 13 months is available on Scale at $0.02 per host per day.

Querying

Everything is stored in tables. The console is a SQL editor with autocomplete over your own schema, and the same tables are reachable over the Postgres wire protocol.

-- slowest endpoints in the last hour, p99 by route
SELECT route,
       count(*) AS n,
       quantile(0.99)(duration_ms) AS p99
FROM traces
WHERE service = 'checkout'
  AND ts > now() - interval '1 hour'
GROUP BY route
ORDER BY p99 DESC
LIMIT 20;

ClickHouse SQL. Standard SELECT, JOIN, CTEs, window functions and the ClickHouse aggregate functions including quantile, uniqExact and topK. Writes are not permitted through the query layer.

Yes. Use the ClickHouse datasource with the read-only credentials from Settings, Query access. Metabase, Superset and anything speaking the Postgres wire protocol also work.

No. There is a concurrency limit of 8 queries per organisation on Team and 32 on Scale, and a 60 second statement timeout. Beyond that, query as much as you like.

What Pulsegrid does not do

There is no browser SDK and no session replay. We cover the server side. If you need RUM, run it alongside from another vendor and correlate on trace ID, which we expose in the response headers.

No uptime probes from external locations. Most teams pair us with a cheap dedicated checker. We will accept the results as an event stream if you want them on the same timeline.

We alert, we do not run the on-call rota. Alerts go to PagerDuty, Opsgenie, Slack or a webhook and the rota lives there. We would rather integrate with the rota tool you already pay for.

Retention tops out at 13 months. If you need seven year retention for audit, export to your own object storage in Parquet, which we support on a schedule, and query it there.