edge --version
p99 14ms · 11MB binary · zero-config single-node, online rebalance to 256 nodes.
A distributed event store for engineers who measure their runtime in microseconds and write their own load tests.
// why edge
Four numbers, taken at saturation.
Every benchmark on this page comes from the public bench suite in the repo. Run it yourself in 90 seconds: edge bench. We publish the failure modes too.
1.2Mops/sec
Single-node throughput
Edge runs a lock-free per-shard event loop. On a 16-core c6i.4xlarge with default tuning, a single node sustains 1.2M write ops/sec and 4.6M read ops/sec on the standard YCSB-A workload.
14msp99
Tail latency, hot-path read
p50 is 0.4ms, p99 is 14ms, p99.9 is 38ms, measured at sustained 80% saturation. Tail latency is the metric that matters in production, so we publish it.
0.6sto rebalance
Cluster topology change
Adding or removing a node triggers an online rebalance. Median time to settle a 32-node cluster is 0.6 seconds, with no read or write availability impact.
11MBbinary
Container image
The Edge daemon ships as an 11MB statically linked binary. No JVM, no Python, no glibc-version dance. The container image, with the binary inside, is 19MB.
// architecture
Lock-free per-shard event loop. Mostly.
Edge is a single binary that can run alone or in a cluster of up to 256 nodes. Each node runs N independent event loops bound to physical cores, with shared-nothing memory. Replication is async by default; tunable to sync per write. The query engine is a MIT-licensed fork of the same vectorized engine that ships in the open-source repo.
// Connect, write, query.
import { Edge } from "edge-tool";
const c = new Edge({ url: "edge://localhost:7080" });
await c.write("events", {
ts: Date.now(),
user: "ks",
kind: "click",
meta: { src: "hero", dest: "/install" },
});
const out = await c.query`
SELECT user, count() AS n
FROM events
WHERE ts > now() - 1h
GROUP BY user
ORDER BY n DESC
LIMIT 10
`;The full architecture document is at docs.edge-tool.example/architecture. 17 pages, no diagrams, written by ks and ml.
// used by
Powering 14B events/day across 47 production teams.
The companies below run Edge on the hot path. Names listed by permission. We do not run a logo wall and we do not write press releases for our customers.
- 01Northpipe
- 02Loomline
- 03Bytebridge
- 04Helsim
- 05Quartzbeam
- 06Mulemark
- 07Grayrock
- 08Daystone
- 09Pivot Reactor
- 10Anseth
Plus 37 teams who asked us not to publish their name. We listened.
// install
One command. No account.
Edge is MIT-licensed at the core. Install it on a laptop, point it at a workload, decide for yourself. The team accounts and the cluster console come in once you want them.
# macOS / Linux
curl -fsSL https://get.edge-tool.example | sh
# Or with brew
$ brew install edge
# Or with Docker
$ docker run --rm -p 7080:7080 edge:latest// pricing
On the page. Not behind a form.
The free tier is full-featured up to 3 nodes. The paid tiers exist because the cluster console, multi-region replication, and the SSO are work we do for the team that buys them.
Solo
$0
For an engineer running it on a laptop or a side project.
- All features, no time limit
- Up to 3 nodes
- Community support on the GitHub Discussions
- No telemetry phoned home
Team
$0.04 / vCPU / hr
For teams running Edge in a production cluster.
- Unmetered nodes, billed per vCPU-hour
- Email and Slack support, 24-hour SLA
- Multi-region replication included
- SSO via OIDC
Hardcore
Talk to us
For teams running Edge at the limits we built it for.
- Custom commitments, custom pricing
- Direct line to the engineering team
- On-site engineering days, when you need them
- BYOC and air-gapped deployments
// documentation
Docs you can read end to end.
The full documentation is one repository, four directories, and a search bar that works without JavaScript. Every code block is copy-paste runnable against the local install.