Welcome — what suslik is, and what isn't here yet

[BennoBaer-dev]

Thanks for stopping by. A suslik is a ground squirrel that stands sentinel and warns its colony of danger, which is roughly what this does. suslik is a small companion service for Frigate: it treats a person's walk across your property as one scenario across several cameras, and only calls a face recognized after several consistent hits, not a single lucky frame. It runs locally (Intel iGPU or NPU via OpenVINO), sits next to Frigate, and changes nothing inside it. The README has the details.

To be upfront: this is an early homelab prototype. The code and a Docker image aren't here yet. This page is where I'm preparing that in the open.

What would help most right now:

• Tell me about your setup: how many cameras, what hardware, roughly how many people you recognize.
• What would make this actually useful for you? Drop ideas or questions here in Discussions.
• Found a specific bug or have a concrete feature request? Open an Issue.

Thanks for the interest so far. That's genuinely what's shaping the direction.

[Tokn59]

Everything is running on 2 NUC's. I have HA, Frigate, EMQX, EspHome, Node Red, Go2RTC, Zigbee2MQTT .. many other containers and now also Suslik the squirrel ! Other hardware : Unifi network equipment, Google Coral TPU. Zigbee network.

[sla004]

Hi Bernd, good news — my setup is back up. Running Frigate on Proxmox 9.2 with NVIDIA RTX 3060 Mobile (12GB) in an LXC container. Currently two Reolink cameras at 1080p, and I'm adding a third (Reolink Doorbell v2) soon, so I'll have 3 cameras total. A handful of people to recognize — family and regular visitors. Happy to test NVIDIA support when you're ready.

[BennoBaer-dev]

Plain Python files, one per language: core/texte/fi.py holds all 2361 strings as key/value pairs, and en.py is the reference with the same keys in the same order. The Finnish file is machine translated, it says so right at the top. For corrections the easiest way for me: quote the key or the English text plus your Finnish wording, here or by mail, and I'll take it over word for word. A release check compares every language file against en.py, so corrected strings can't drift out of sync later.

One more thing I'd like to know: how good or bad the Finnish is overall. If it makes systematic mistakes, wrong tone, wrong grammar patterns, that kind of thing, tell me straight. Then I'd rather redo the whole translation than patch it string by string.

[sla004]

I went through fi.py against en.py, first with scripts over all 2361 keys and then by reading a sample and targeted searches by hand.

Overall verdict: the Finnish is good, so I would not redo it. Keys, placeholders and HTML tags are intact, the register (informal "you") is consistent and most sentences read naturally. The problems are systematic in three places:

Terminology is inconsistent. The same concept is translated differently in different places (settings/configuration, stage, refresh, match, gallery/catalogue/library, train/learn).
Some technical words were left in English (score, Easy/Expert, Host, "User key:", "feature norm", health).
Occasional literal machine phrasing, e.g. "still running" rendered as "still computes".

A glossary plus about 20-40 targeted fixes should be enough. If you prefer to regenerate, feed the glossary below to the translation agent.

Glossary (decided):

score → tulos
stranger (Fremd) → tuntematon henkilö
train / learn → opeta / opettaa (one verb everywhere)
gallery / catalogue / library → kirjasto for all three
health → tila
feature norm → ominaisuusnormi
Easy / Expert → Helppo / Asiantuntija
Host → palvelin
User key: → Käyttäjäavain:
settings / configuration → asetukset everywhere ("Advanced" stays "Lisäasetukset")
stage → vaihe, refresh → päivitä, threshold → kynnys, match → osuma, reference image → viitekuva
Kept as is: token, Vision, Frigate, Video, Live, Beta

Specific fixes (key → Finnish):

antwort.lernlauf_resume_aktiv → yksi tämän ajon vaihe on vielä käynnissä
anwesenheit.legende_weg → järjestelmä oli käynnissä, ketään ei vahvistettu
leer.tag → Ei havaittuja kasvoja tälle päivälle
vision.galerien.satz → Kirjasto voi olla vain henkilöillä, joille on opittu kehomalli
lernen.karte.metrik_kurz → tulos {score}
vision.galerien.knopf_auffrischen → Päivitä
systemstat.live_knopf.tip_aus → Päivitä {takt} sekunnin välein … Myös sivulta poistuminen lopettaa sen.
visiontest.koerper.nichts → mitään ei arvioitu
konfiguration.support.einmal_hinweis → … vain kerran … sen jälkeen näkyy vain •••
auftritte.zuweisen.satz → "junk stays behind" should read: roskakuvat jäävät pois (currently "kelvoton jää jäljelle", which means the opposite)
js.ref.batch_alle_frage → Nämä ovat KAIKKI henkilön {person} {n} viitekuvaa. (word order)
livekalib.regler_k_prosa → replace "käydä silti äänestä" with "vaikuttaa silti tunnistukseen"
ui.modus.easy / ui.modus.expert → Helppo / Asiantuntija
ui.fuss.health → tila
benachrichtigungen.pushover.label_user → Käyttäjäavain:
benachrichtigungen.mqtt.label_host → Palvelin: (and "Tämä host" in vision.hinweis.host_satz → "Tämä palvelin")
livekalib.katalog.regler_n, livekalib.ohne_norm, livekalib.pruefen.prosa, js.import.fertig_wiz, js.import.fertig_ges, setupwiz.import.satz: "Feature-normi" → ominaisuusnormi
The 18 rows containing "score" → "tulos"; the ~55 catalogue/library/gallery rows → "kirjasto" (needs inflection care, e.g. galleriaan → kirjastoon)

One structural note (not a translation issue): 176 rows use "{n} + partitive" ("{n} kuvaa", "{n} tapahtumaa"), which is wrong for n = 1. You mentioned t_n for later, so I assume that is planned.

I can send a full list with every affected row as key → English → Finnish if that is easier for you to apply, just tell me.

[BennoBaer-dev]

Thank you so much for this. It's exactly the review I was hoping for: the glossary settles the terms, and the fixes I can apply one to one. I'm building all of it into the next fix release, hopefully out tonight. The plural forms for n = 1 you guessed right, that's the planned counting rework. And on your offer: if the full row list isn't too much work for you, I'll gladly take it and apply it the same way.

The usual note: I write in German and let an LLM translate, my own English is not up to a text like this.

[sla004]

Thank you, and good luck with the release tonight. The full list is attached.

Files

fi_corrections.py: a Python dict CORRECTIONS = {key: new Finnish}. Each value replaces the current value in fi.py one to one. 152 rows.
fi_corrections_review.md: the same rows with the English text, the old Finnish and the new Finnish, for checking.

I checked with a script: all keys exist in en.py, and placeholders, HTML tags, href and contents and edge spaces are unchanged.

What is in it

All decisions from the glossary: score → tulos, stranger → tuntematon henkilö, train → opeta/opetus, gallery/catalogue/library → kirjasto, health → tila, feature norm → ominaisuusnormi, settings/configuration → asetukset, Easy/Expert → Helppo/Asiantuntija, Host → palvelin, User key → Käyttäjäavain.
The specific fixes from my first message.
"Vision" and "token" stay as they are.

Points to check

"oppimisajo" (learning run) and "oppii" are unchanged on purpose. Only the "train" words (koulutus) became "opetus". If you want "learning run" changed too, tell me.
Both "unknown" and "stranger" are now Finnish "tuntematon". Rows for unknown visitors still say "tuntematon vierailija", and rows for strangers now say "tuntematon henkilö". They should stay distinguishable in context, but please look at lernen.karte.unbekannt, which is now just "Tuntematon henkilö".
In vision.modell.antwort_satz the words residents and strangers were left in English inside the sentence. I translated them (asukkaat, tuntemattomat henkilöt). If the table column headers in the UI are hard-coded in English, they need to match.
Feature norm: only 3 rows contain it in Finnish (livekalib.katalog.regler_n, livekalib.ohne_norm, livekalib.pruefen.prosa). I said six keys before, which was wrong.
In personwizard.kontrolle.schalter_satz the link /konfiguration is unchanged. Only the visible label changed to "Asetukset".
fi_corrections_review.md
fi_corrections.py

[BennoBaer-dev]

Thanks Simo, the list and the review file make this easy for me. It won't make it into tonight's release though. The full run with QS and the privacy checks takes almost 4 hours and it's already in its last stage. Your changes go in tomorrow, and I'll look at the points you flagged then. And thanks for the good luck wish, the release is running while I write this.

The usual note: I write in German and let an LLM translate, my own English is not up to a text like this.