# Support policy

What Kwker supports - versions, platforms and products - how to get help, when you hear back, and how security problems are handled.

Web page: https://kwker.io/docs/?p=support - every Kwker documentation page for AI assistants: https://kwker.io/llms.txt

Kwker is supported through the documentation, the support form at [kwker.io/support](https://kwker.io/support/) and
[mail@kwker.io](mailto:mail@kwker.io). This page says what is covered and what to expect.

## Versions

| Version | Fixes |
|---|---|
| The newest 0.1.x release | Bug and security fixes, as 0.1 patch releases |
| Older minor versions and development snapshots | None: upgrade to the newest release |

A patch release never breaks a caller, so moving to the newest 0.1.x is always a drop-in upgrade. What "never breaks"
means for each package is in [Compatibility](https://kwker.io/docs/md/compatibility.md).

## Platforms

| Platform | Level |
|---|---|
| Linux x86-64 (AVX-512, AVX2 and older CPUs) | Supported |
| Linux ARM64 | Experimental |
| macOS on Apple silicon | Experimental |
| Windows x64 and ARM64 | Experimental |

Supported means every package is tested on every release and its speed is measured and published. Experimental means
the packages are tested on every release and give the same results as on Linux x86-64, but options and speed may change
in a patch release. The install pages list what each platform has: [Linux ARM64](https://kwker.io/docs/md/install-linux-arm64.md),
[macOS](https://kwker.io/docs/md/install-macos.md), [Windows](https://kwker.io/docs/md/install-windows.md).

The torch.compile backend, the ONNX Runtime provider, the DuckDB extension and the Polars plugin are experimental on
every platform. `kwker.EXPERIMENTAL` lists the experimental parts.

## Getting help

1. Check [Troubleshooting](https://kwker.io/docs/md/troubleshooting.md) and [Known limitations](https://kwker.io/docs/md/limitations.md).
2. Run `python -m kwker doctor` (or `bin/kwker-doctor` from the C / C++ package): it shows the engine in use and any
   warnings.
3. Write to us through the [support form](https://kwker.io/support/) or [mail@kwker.io](mailto:mail@kwker.io). For a
   bug, attach a support bundle (`python -m kwker support-bundle`) and a small program that shows the problem: the
   input size, the key type and a seed or the data.

## What counts as a bug

- A wrong result: a sort, selection or key-value result that differs from the documented order, or a model output
  outside its precision's documented error bound. These come first.
- A crash, a hang or memory use beyond the documented limits.
- A confirmed slowdown against an earlier Kwker release on the same machine.
- An install path in these docs that does not work as written.

Speed against another library on your workload is a question we are glad to look at, but it is not a defect.

## When you hear back

We answer on business days and aim to reply within five of them. There is no 24/7 support and no service-level
agreement yet. Paid support plans with response times per tier are not offered yet; Commercial and OEM license holders
use the same form (choose "Licenses and pricing").

## Security problems

Report a suspected vulnerability privately to [mail@kwker.io](mailto:mail@kwker.io), not through the form or a public
issue. The [security policy](https://kwker.io/security/) says what to include. Our targets: an acknowledgement within 7
days, an assessment within 14 days, and a fix or mitigation for a confirmed memory-safety issue within 30 days.

## Related

- [Troubleshooting](https://kwker.io/docs/md/troubleshooting.md): doctor warnings, common questions, the support bundle.
- [Compatibility](https://kwker.io/docs/md/compatibility.md): what stays stable within 0.1.x.
- [Licensing](https://kwker.io/docs/md/licensing.md): which license applies to you.