feat(desktop): provider/model manager

[wildminder]

What does this PR do?

Adds a Provider / Model Manager settings view to the desktop app. Users can browse all configured providers (built-in + custom), toggle per-model visibility, add/edit/delete custom OpenAI-compatible providers, activate/deactivate providers, and discover or refresh model catalogs - all from a two-pane Settings UI. The backend gains model-discovery and provider-filtering endpoints to support it.

Related Issue

N/A (new feature)

Type of Change

• ✨ New feature (non-breaking change that adds functionality)
• ✅ Tests (adding or improving test coverage)

Changes Made

Desktop — new Settings view (apps/desktop/src/app/settings/)

• provider-model-manager.tsx — two-pane container (provider nav + model list), search, deep-link via ?pmprovider=<slug>
• provider-manager-nav.tsx — left pane: ARIA listbox, arrow-key nav, per-provider active-model count pill, enable/disable switch
• provider-model-list.tsx — right pane: per-model Switch toggles, capability badges, letter grouping (opt-in), sort A–Z/Z–A, active-model bold emphasis
• provider-form-dialog.tsx — add/edit custom provider modal (name, base_url, api_key, api_mode, models); auto-generates provider ID from name
• provider-configure-panel.tsx — edit-credentials dialog for built-in providers
• model-add-dialog.tsx — manual model addition for custom providers
• provider-grouping.ts, model-list-utils.ts — pure helpers (letter grouping, capability color meta, discover-button label)
• use-provider-catalog.ts, use-provider-config.ts, use-provider-visibility.ts — TanStack Query + nanostore hooks

Desktop — lib / store / types / i18n

• lib/custom-provider-config.ts — pure-TS config helpers (upsert/remove/enable, name normalization, models list↔dict)
• store/model-visibility.ts — extended with provider-scoped helpers
• types/hermes.ts — enabled? on ModelOptionProvider
• i18n/{en,ja,zh,zh-hant,types}.ts — providerManager namespace (~30 keys × 4 locales)
• hermes.ts — getGlobalModelOptions({ includePricing, includeUnconfigured, explicitOnly, refresh }), discoverProviderModels()

Backend

• hermes_cli/inventory.py — build_model_options_payload(include_pricing=…), provider enabled flag, model display names, discover_provider_models()
• hermes_cli/web_server.py — GET /api/model/options (include_pricing param), POST /api/model/discover
• tui_gateway/server.py — model.options respects model.disabled_providers + custom enabled: false

Tests

• 5 new Python test files (test_inventory_discover.py, test_inventory_model_display_names.py, test_inventory_options_flags.py, test_inventory_provider_enabled.py, test_web_server_model_discover.py)
• 13 new/extended TS test files covering all components, hooks, and pure helpers

Toolchain

• apps/desktop/package.json — vitest pinned to ^3.2.7 (4.1.x crashes on Node 26)
• apps/desktop/vitest.setup.ts — localStorage polyfill for jsdom

How to Test

1. Launch the desktop app (_run_desktop.bat or hermes dashboard), open Settings → Provider Manager.
2. Verify the provider list renders with model-count pills; select a provider and toggle models on/off — confirm toggles persist across remount.
3. Click Add provider, fill in a custom OpenAI-compatible endpoint (e.g. http://localhost:11434/v1), save — confirm it appears in the nav with all models disabled by default.
4. Click Discover models on the custom provider — confirm models populate. Toggle the provider off/on — confirm models hide/show.
5. Run cd apps/desktop && npx tsc -p . --noEmit && npx tsc -p tsconfig.electron.json --noEmit — both exit 0.
6. Run scripts/run_tests.sh tests/hermes_cli/test_inventory_discover.py tests/hermes_cli/test_inventory_options_flags.py tests/hermes_cli/test_inventory_provider_enabled.py tests/hermes_cli/test_inventory_model_display_names.py tests/hermes_cli/test_web_server_model_discover.py -q — all pass.

Checklist

Code

• I've read the Contributing Guide
• My commit messages follow Conventional Commits (feat(desktop): …)
• I searched for existing PRs to make sure this isn't a duplicate
• My PR contains only changes related to this fix/feature (no unrelated commits)
• I've run pytest tests/ -q and all tests pass
• I've added tests for my changes (13 TS test files + 5 Python test files)
• I've tested on my platform: Windows 10

Documentation & Housekeeping

• N/A — no user-facing docs changes (feature is self-documenting via i18n UI)
• N/A — no new config.yaml keys (uses existing custom_providers / model.disabled_providers)
• N/A — no architecture/workflow changes to AGENTS.md
• I've considered cross-platform impact — the feature is pure Electron/React + Python backend; no OS-specific primitives
• N/A — no tool schema changes

Screenshots / Logs

[teknium1]

Thanks for the substantial Desktop UI and test work. Current main has moved custom-endpoint management onto a keyed providers: schema, so this needs a focused rework before it can safely land.

Problems

• apps/desktop/src/lib/custom-provider-config.ts:204 writes a custom provider API key into the config record. Current main deliberately keeps these keys in .env and stores key_env instead (hermes_cli/web_server.py:7189-7210).
• apps/desktop/src/lib/custom-provider-config.ts:164-214 only manages legacy custom_providers, while current v12 custom endpoints use providers: (hermes_cli/config.py:1508-1571; hermes_cli/web_server.py:7133-7223). This can omit current entries or create duplicate representations.
• apps/desktop/src/app/settings/use-provider-catalog.ts:26 uses REST catalog data, but the PR adds enabled annotation only in the TUI JSON-RPC path. The proposed model.disabled_providers state at custom-provider-config.ts:252-268 also differs from main's runtime-supported providers.<name>.enabled configuration (hermes_cli/config.py:2850-2872).

Suggested changes

• Rebuild persistence and enablement on the current providers:/key_env lifecycle, and expose one authoritative enablement state through the REST catalog.
• Add end-to-end coverage for v12 provider edit, credential storage, disable, and re-enable.

Automated hermes-sweeper review.