Any metric · across versions

Every number your build prints, tracked across versions.

Size, RAM, latency, accuracy, cost — any metric, pushed from CI in one call. See what changed between versions, sliced by product and target, and decide with data.

Join the first five teams →or, for open source

Drift = the slow creep of your numbers between releases — not ML data-drift monitoring.

Flat price per org — never per seat. We never see your source or your binary.

Interactive demo · sample data
v2.5.0 +10.9% vs v2.4.0 · tap a point for details
You already do this

A chat thread where someone pastes the numbers after each release.

A 917-line script that prints the metrics table.

A model-zoo README with size, latency and accuracy per model.

A spreadsheet of RAM × accuracy × latency per target.

A GitHub issue that has been the dashboard for six years.

Every team that cares about these numbers has already built a worse version of this. It works — until keeping it current is someone’s job, and nobody sees the regression until a release ships it.

How it works

One HTTP call from wherever you build — CI, a nightly job, your laptop. No agents, no infra.
01 · push
Push any number, from anywhere

Anything your build already prints — size, RAM, latency, accuracy, cost per build — tagged with the version and target. See the docs →

$ curl -X POST https://api.driftmetrics.watch/v1/push \
    -H "Authorization: Bearer $DRIFT_TOKEN" \
    -d '{"tags": {"version": "v2.5.0", "platform": "linux-x64"},
         "metrics": [{"name": "binary_size", "value": 2.14}]}'→ deltas: binary_size +10.9% vs v2.4.0
02 · see
History, diffs and slices

Every metric’s history across versions or over time. Compare any two versions side by side, and filter or group by the tags you push — product, target, branch.

03 · share
Show what changed

Save the views you check every release, and send a share link — password protected if you like — so anyone can see what changed between versions.

Coming soon: Slack alerts · PR comments when a metric deviates · blocking PRs on a budget

Use cases

Whatever you ship, push the numbers that drive your decisions.
push
model size, peak RAM, latency per inference, accuracy
slice by
model, quantization, runtime, device
decide
Pick the model that fits the device, and see which release made it slower.

We never see your source or your binary.

Driftmetrics ingests numbers — a metric name, a value, a few tags. Your build produces them and your CI keeps everything else. There is no artifact upload to review and nothing to parse on our side.

  • no source
  • no binaries
  • no artifacts
  • no build logs

Flat price per org. Never per seat.

Everyone reads the numbers, so charging by the head would tax the one thing that makes it work.
Open source
Freea few projects, hand-picked
  • Gating and PR checks included
  • Unlimited history and metrics
  • I wire it up for you, personally
Team
$40per month, per org
  • Private projects, unlimited seats
  • Product / component / version faceting
  • Budgets, Slack, shareable reports
Business
$150per month, per org
  • Unlimited projects, SSO
  • Self-hosted for regulated teams
  • Priority support

Alpha pricing — the first five teams keep three months free.

Open source

Maintain something open source? Reach out.

I take on a few projects at a time, free, and set them up myself: send me the repo and I’ll set up your dashboard and open the pull request that pushes your numbers from CI — you review it like any other PR. I pick them by hand, so tell me what you’d want to track.

Repo you maintain or contribute to

A handful at a time — I reply either way, within a day.

Be one of the first five teams.

I’m onboarding 5 teams personally — free for 3 months — to track the numbers they care about across every release.

Get set up personally — free for 3 months

Work emails only, please — it helps me line up the right early conversations.

Prefer email? Write to me directly →