---
id: fleet/settings-to-choose-which-server-each-feature
title: "Settings to choose which server each feature uses"
theme: fleet
status: open
priority: urgent
due: 2026-08-22
created: 2026-08-21
date_source: stated
order: 0
---

**Settings to choose which server each feature uses.** Wanted 2026-08-21, **needed 2026-08-22** — the
owner named it urgent and named the case: **ReClip, for saving videos.** Now that there are several
servers, each functionality needs to be pointed at one deliberately, from Officer's settings, rather
than always resolving to whatever is local.

**The blocker is concrete and small.** `api/file-browser/router.ts:1271` reads

```ts
const RECLIP_BASE = process.env.RECLIP_URL ?? 'http://localhost:8899';
```

— a module-level constant, read once at import, from an environment variable. So every install can only
ever reach its OWN ReClip, there is no way to say "download this on `omega`", and changing it means
editing `.env` and restarting rather than a setting. Five hosts hold a platform checkout
(`docs/servers.md`); three run it.

The generalisation the owner asked for is the setting being **per feature**, not one global "which
server am I". Different functions belong on different machines — the box with the disks is not the box
with the GPU is not the box on the fast line — so this is a routing table, one row per capability.

Worth knowing before designing it: the platform already has per-server registries for a plugin
(Jellyfin keeps servers with write-only access tokens; offscale keeps a server picker), so there is a
shape to copy rather than invent. And `docs/multi-server-push.md` is the same question arriving from the
other direction — a payload that does not say which server sent it — so whatever names a server here
should be the identifier that design settles on.