What it is not
Written down so you can rule us out in four minutes instead of a fortnight. None of these are on a roadmap; a few of them we have actively decided against.
- Not a scheduler. Slatepipe never starts your work. If the thing you need is dependency resolution, backfills or a retry policy, you need a scheduler and you will still have one after installing this.
- Not a log store. We keep the last 64 KB of standard error from each run, which exists for triage, not for searching. Grepping six months of application logs is a different product with a different storage bill.
- Not service monitoring. There is no concept of a request, a latency percentile or an uptime check. A long-lived service that never exits produces exactly one unfinished run here, forever, and that is the correct answer rather than a bug.
- No column-level lineage. We record the counts a job reports at its own step boundaries. Which column came from which upstream table is something we cannot see without parsing your SQL, and we are not going to pretend to parse your SQL.
- Not a data quality framework. Row counts and durations compared against history, nothing else. Uniqueness, referential integrity and distribution checks belong in the job itself, where they can stop the pipeline before it publishes.
- No roles or per-team views. One read-only interface, one admin token, no accounts. If an auditor needs to know who looked at a dashboard, this design will fail that audit and we would rather say so here.
- No hosted version. You run it, you keep the data. Partly because the data is usually more sensitive than people expect once you see command lines in it, and partly because three people cannot operate someone else's production database at a price either of us would like.
- No streaming. The unit of everything here is a run with a start and an end. A continuous consumer has neither, so the charts would be empty and the alerts meaningless.
- No agent for Windows. The wrapper depends on process-group handling we have only implemented for Linux. The HTTP endpoint works from anywhere, so a PowerShell script can still report runs; it just has to do its own retries.
- No support promise on the free tier. Apache-2.0 means you may run it forever with no payment and no feature gate, and it also means nobody owes you an answer by Tuesday. Paying for support is what buys the Tuesday.
If three of those matter, we are the wrong tool and this page did its job. If none of them do, the install page is fifteen minutes long and the whole thing is two binaries you can delete.