Models are browsed by source, not one list with Source and Capability filters
メンテナーはふだん 1 日以内に返信
@ybai08 がすでに取り組んでいます。
2026年10月16日 から。
評価
- 難易度
- 4/5
- 見積もり時間
- 3〜5日
- 初心者へのやさしさ
- 45/100
- issue の種類
- 機能追加
- 明瞭さ
- 明確に書かれている
- 活発さ
- 活発
- 技術スタック
- react, typescript
調査の方向性
Read docs/proposals/model-catalog.md first, especially the Screen section and step 6; then inspect the named Add model and per-source group components under surfsense_local/frontend/src/features/models/. Start by writing a failing filter-behavior test, using the listed settings and download-model tests as guides. Done means the scoped part of the proposed unified list and filters is covered, with relevant docs updated; run pnpm test and python scripts/check_docs.py.
索引モデルが issue の本文から書いたものです。
説明
The gap
docs/architecture/local-models/catalog.md lists this under Known gaps:
Browsing is still split by source, a catalog on the Add model page and one group per server, not the one list with Source and Capability filters the proposal describes.
Why it matters
You have to know where a model came from before you can look for one. Answering "what can I use for image editing?" means reading This computer, then each connected server's group, then opening Add model and reading the catalog there, then the search box — four places, and again for each of the five slots. The one question a person actually has, "what can do this job", is the one the layout cannot answer.
It is also the last piece holding several docs in a transitional state. This page opens by saying the one screen with Source and Capability filters is still to come and that the page moves to model-catalog/local.md when it does; connections.md says the same from its side. Until this ships, three architecture docs describe an interim shape.
Where
docs/proposals/model-catalog.md— read this first. The screen section and step 6 of Order of work are the design; this issue is that step, not a fresh one. Steps 1 to 5 have shipped, which is why the backend already returns everything the filters need.surfsense_local/frontend/src/features/models/add-model/add-model-options.tsx— the two-section Add model page.surfsense_local/frontend/src/features/models/your-models/local-models-group.tsxandremote/models/server-models.tsx— "This computer" and the per-server groups, the split itself.surfsense_local/frontend/src/features/settings/models/— the five per-slot settings pages that assemble those groups.- Reuse rather than rewrite: a row already carries
selectable_for,types,knownandengine, so the Capability filter is a filter over data the API already sends, andmodules/llm/selectable.pyis the one rule for which slots a model fills.ModelCard,BuildActionandInstallProgressare the row parts; they should survive the rework.
What done looks like
- One list of models with a Source filter (All, Local, Remote) and a Capability filter offering each type that has rows, plus Unknown when any row is unknown — as the proposal specifies.
- Hugging Face search stays its own box. That is deliberate and the proposal repeats it: what the user types goes to a third party, and it carries the egress question.
- This is large enough to split, and splitting it is welcome. Say in the PR which part of step 6 it is and what is still to come.
- The docs move with the work, and that is a real part of it: this page becomes
model-catalog/local.md,connections.mdkeeps storage, keys, the probe and the runtime, and the proposal is deleted when the last piece lands. Do only the part your change earns.
What not to change
Do not change what the API returns to make the screen easier. The rows are the finished part of this design — the backend decides and the screen renders, computing no fit, no default, no recommendation and no selectability. A filter that recomputes any of those in the renderer is the failure mode this whole design exists to avoid.
How to verify
The per-slot settings pages are covered by chat-models-settings.test.tsx, image-models-settings.test.tsx, audio-models-settings.test.tsx and their siblings, and the catalog itself by download-chat-models.test.tsx. The repo is TDD: write the failing test for each filter behaviour first.
cd surfsense_local/frontend
pnpm test
cd ../.. # this touches docs/, so check them from the repo root
python scripts/check_docs.py
Before you start
Read CONTRIBUTING.md, the model catalog proposal, this page's local model catalog doc, the connections doc and ADR 0014. React work follows the frontend-workflow skill, shadcn components the shadcn skill after it, and interface strings the translate skill, all in AGENTS.md. This is a large change: comment here with what you plan to take on before you start, so it can be split rather than collided with. Open your PR against dev, not main.
- 主要言語
- Python
- スター
- 16.3k
- フォーク
- 1.6k
- 平均マージ
- 17時間 40分
- マージ済み PR(30日)
- 290
環境構築
このプロジェクトには開発コンテナ、Dockerfile、コントリビューションガイドがありません。まず README を読み、一般的な手順ははじめてのコントリビューションガイドを参照してください。
はじめの一歩
- issue を最後まで読み、次にプロジェクトのコントリビューションガイドを読みます。
- 着手することを issue にコメントします — 二人が同じ作業をするのを防げます。
- リポジトリをフォークし、ブランチを切って変更します。
- issue 番号を参照したプルリクエストを送ります。
MODSetter/SurfSense のほかの issue
-
Security Scan: stop flagging pinned hashes and test fixtures on release PRs対応中かも @karthikeya1220 が 14 日前に担当しました。 オープンcode quality security
難易度 2/5 1〜3時間 初心者へのやさしさ 78/100
MODSetter/SurfSense#2225 · コメント 1 件 ·
メンテナーはふだん 1 日以内に返信
-
難易度 2/5 1〜3時間 初心者へのやさしさ 73/100
メンテナーはふだん 1 日以内に返信
-
backend bug desktop
難易度 4/5 3〜5日 初心者へのやさしさ 48/100
MODSetter/SurfSense#2366 · コメント 5 件 ·
メンテナーはふだん 1 日以内に返信
-
[bug] 直接把桌面上的这个文件拖进去オープンbug
難易度 3/5 1〜2日 初心者へのやさしさ 45/100
メンテナーはふだん 1 日以内に返信
-
難易度 1/5 1時間未満 初心者へのやさしさ 55/100
メンテナーはふだん 1 日以内に返信
MODSetter/SurfSense の issue をすべて見る
似ている issue
-
難易度 2/5 1〜3時間 初心者へのやさしさ 65/100
メンテナーはふだん 1 日以内に返信
-
[oblt-aw][autodoc] Document obs-aw-automerge contract and fix automerge doc inconsistencies対応中かも @github-actions が今日担当しました。 オープン
難易度 2/5 1〜3時間 初心者へのやさしさ 76/100
メンテナーはふだん 1 日以内に返信
-
admission-deferred priority: medium status: ready work-class: scoped
難易度 2/5 1〜3時間 初心者へのやさしさ 70/100
melodic-software/claude-code-plugins#10887 · コメント 4 件 ·
メンテナーはふだん 1 日以内に返信
-
backend bug good first issue
難易度 2/5 1〜3時間 初心者へのやさしさ 70/100
PoCInnovation/Distribox#255 ·
メンテナーはふだん 1 日以内に返信
-
難易度 2/5 1〜3時間 初心者へのやさしさ 72/100
google/langextract#598 ·
メンテナーはふだん 1 日以内に返信