feat(ai): rework AI configuration into a unified view

[ndoschek]

What it does

Consolidates AI configuration into a single master–detail view with flat category
navigation (General, Providers & Models, Model Aliases, Agents, Prompts & Skills,
Variables, Tools, Token Usage, MCP Servers), retiring the legacy per-tab widgets and the
scattered AI entries in the Settings UI.

• Adds the AiConfigurationService seam and the public AiConfigurationCategory
  contribution point with an ordered, lazily-wired registry (ai-core, ai-core-ui), so
  the view reads/writes AI preferences workspace-trust-aware and stays insulated from how
  they are stored.
• Adds shared, DI-free page primitives: sections, list rows, settings rows and controls
  (toggle/select/number/array), a gear context menu (Copy Setting ID / Reset Setting),
  status badges, and a filterable collapsible list.
• Renders agents (prompt variant sets, language models, notifications, capabilities),
  providers & models, model aliases, prompts, skills & slash commands, variables, tools and
  MCP servers as consistent list/detail pages; complex preferences defer to settings.json.
• Hides AI preferences from the Settings UI and redirects ai-features.* deep-links to the
  view; shows the category → item path as real Theia breadcrumbs.
• Providers & Models discovers provider blocks by a positive aiModelProvider schema
  marker (no denylist), so unmarked feature areas fall through to General's catch-all.

Refs eclipsesource#310, eclipsesource#311, eclipsesource#313,
eclipsesource#314, eclipsesource#316.

Resolves GH-17649.

flowchart LR
    Cat["Config areas, each a category
    (built-in + contributed by extensions)"]
    View["AI Configuration view
    tree on the left, detail page on the right"]
    Prim["Shared UI building blocks
    (rows, controls, sections)"]
    Svc["AiConfigurationService
    single read/write point"]
    Store["Settings storage
    (preferences + per-agent settings)"]

    Cat -->|registered and listed in| View
    View -->|detail pages built from| Prim
    Prim -->|read and write through| Svc
    Svc --> Store

Loading

How to test

• Open the AI Configuration view from its entry points, move between categories and items, and confirm the breadcrumbs follow.
• Use the tree search to find a setting and confirm it navigates to the owning category and row.
• On the General page, edit AI feature preferences of each control type and confirm they persist and can be reset.
• On Providers & Models, confirm real providers appear as their own blocks (and non-provider feature areas do not), and that provider settings edit and persist.
• Open Model Aliases, Agents, Prompts & Skills, Variables, Tools, MCP Servers and Token Usage, and exercise each page's main actions (view, edit, add or remove, reset where applicable).
• Confirm AI preferences are managed in this view and that AI settings entry points and deep-links route here (e.g. via commands, toolbar items, links in the AI Chat welcome pages).
• Enable and disable AI features and confirm the view reacts correctly.
• The AI Configuration view got a default keybinding (Alt+a) and also an entry in the bottom left gear menu ("Manage").

Follow-ups

• Automatic model discovery (AI Provider Model List Management eclipsesource/theia#318): when a provider exposes a model-list endpoint, discover its available models automatically instead of requiring them to be configured by hand.
• Scope handling ([AI Config] Scope handling eclipsesource/theia#312): the view currently edits a single scope (user scope). Complete the scope UX, i.e. selecting and switching user, workspace and folder scopes, and reflecting per-scope state consistently across all rows.
• Simple / Pro switch ([AI Config] Simple / Pro toggle and placeholder eclipsesource/theia#315, [AI Config] Simple-mode config contents eclipsesource/theia#317): add a simple vs. advanced ("Pro") mode so the view can present a curated, minimal set of settings by default and reveal the advanced ones on demand.
• Custom renderers for complex settings: structured settings that currently fall back to "Edit in settings.json" (e.g. Anthropic custom models) should get dedicated in-view editors instead of requiring the user to edit raw JSON.

Breaking changes

• This PR introduces breaking changes and requires careful review.

Adopter-facing notes are in doc/Migration.md (AI Configuration View): the per-tab
config widgets and their WidgetFactory registrations are removed, @theia/ai-ide and
@theia/ai-mcp now depend on @theia/ai-core-ui, and custom AI-config tabs must be
re-implemented as AiConfigurationCategory contributions. Stable entry points
(aiConfiguration:open, aiConfiguration:openTools, chat toolbar button) are unchanged.

Attribution

Review checklist

• As an author, I have thoroughly tested my changes and carefully followed the review guidelines
• User-facing text is internationalized using the nls service (for details, please see the Internationalization/Localization section in the Coding Guidelines)

Reminder for reviewers

• As a reviewer, I agree to behave in accordance with the review guidelines

[ndoschek]

@EclipseSourceAI

[EclipseSourceAI]

Note

Autonomous AI review.

This review was done by an AI agent and therefore may contain mistakes. Feel free to ignore any comment you disagree with. Noting why helps, since replies are read and taken into account in follow-up reviews.

Resolving all AI comments does not lead to an automatic approval. A maintainer still needs to review and sign off on the overall architecture and design.

To get an updated review after pushing changes, a maintainer may re-request a review from this account.

Running in Eclipse Enclave, submitted via review-guard-mcp

Consolidates the AI configuration tabs into one master/detail view behind a new public AiConfigurationCategory contribution point in @theia/ai-core-ui, hides ai-features.* from the Settings UI, and reroutes every AI-settings entry point here. Large but well-structured: lint is clean, all 1046 tests in the touched packages pass, and driving all 50 tree nodes in the browser produced no blank pages and no console errors, with edits persisting across reload.

Suggested focus for a maintainer:

• ai-configuration-category.ts is the new public API. highlight.subId, ctx.filter and ctx.scope are inert today, which is harder to walk back after adopters see them.
• HideAiPreferencesContribution carries the whole "AI prefs are no longer in Settings" cutover on a one-shot setTimeout(0) pass.
• initializeLayout now auto-opens the view on every fresh layout, so all adopters' first start lands on it even with AI disabled. That looks like a product decision worth confirming.
• The getOwnedPreferenceIds() ownership contract is what keeps General's catch-all from double-rendering, and three categories declare it with string literals instead of the existing constants.

Two confirmed defects on the General page: the hero row lacks the data-ai-config-row-id anchor its own search item deep-links to, and ai-settings-row-id has no CSS rule, so the raw preference id renders next to the title at heading weight. Search-result scroll/flash is also intermittent. Details inline.

[EclipseSourceAI]

Note

Autonomous AI review.

This review was done by an AI agent and therefore may contain mistakes. Feel free to ignore any comment you disagree with. A thumbs-down reaction on a comment marks it as rejected for follow-up reviews. Noting why in a reply helps, since replies are read too.

Resolving all AI comments does not lead to an automatic approval. A maintainer still needs to review and sign off on the overall architecture and design.

To get an updated review after pushing changes, a maintainer may re-request a review from this account.

Running in Eclipse Enclave, submitted via review-guard-mcp

Went through d5faabaef (the "address review feedback" commit) against every thread from my previous review. All nine fixes described in that commit's message check out in the actual code: the hero row anchor, disabled catch-all rows, the deterministic scroll-and-flash retry with the new centerInBody (avoids the scrollIntoView ancestor-scroll bug), dropping the auto-open-on-fresh-layout initializeLayout, the onDidChangeSchema subscription in HideAiPreferencesContribution (with good new tests), routing the token-usage deep link through AI_SHOW_SETTINGS_COMMAND with argument forwarding on both the ai-core and ai-ide sides, using the existing preference-id constants, dropping the dead ctx.filter/highlight.subId/search-dropdown helpers, and reworking ModelsConfigurationCategory onto CollectionCategoryRenderer. Nice, focused fix commit with tests added where it mattered.

One loose end: a reply on the ai-core-ui/package.json thread claimed a stale JSDoc comment (about which context menu the gear button opens) was fixed as a side effect, but the comment is unchanged, flagged inline.

I wasn't able to get a full local build going in this environment (native module/postinstall issues unrelated to the PR), so this pass is code-review only, no live UI check this time.

These previous comments can be resolved as they are now handled:

• Enable AI features row never got its search deep-link anchor
• raw preference id rendered unstyled next to the hero title
• catch-all rows stayed editable while AI features were off
• tool preference ids hardcoded instead of using existing constants
• duplicate nls key registered with different English text
• AGENT_SETTINGS_PREF/LANGUAGE_MODEL_ALIASES_PREFERENCE not reused for the preference id
• agent/notification descriptions duplicating nls keys with different text
• chat-input-widget hardcoded the ai-ide command id instead of the generic one
• hide-ai-preferences only covered schemas registered in the initial sync pass
• pendingHighlight consumed before the target row was mounted
• view opened as an active tab on every fresh layout
• highlight.subId/ctx.filter dead contribution-point API
• unused search match()/MAX_RESULTS helpers
• unused limitedLabel/limitedDefault props on AiSessionLimitControl
• ModelsConfigurationCategory reimplementing CollectionCategoryRenderer

I can't resolve them myself as I would need write permission on this repository.

[EclipseSourceAI]

Note

Autonomous AI review.

This review was done by an AI agent and therefore may contain mistakes. Feel free to ignore any comment you disagree with. A thumbs-down reaction on a comment marks it as rejected for follow-up reviews. Noting why in a reply helps, since replies are read too.

Resolving all AI comments does not lead to an automatic approval. A maintainer still needs to review and sign off on the overall architecture and design.

To get an updated review after pushing changes, a maintainer may re-request a review from this account.

Running in Eclipse Enclave, submitted via review-guard-mcp

Only one commit landed since my last pass: 277eaa49e ("test(ai): cover the AI configuration contracts the view relies on"). It adds regression tests for exactly the DOM-boundary defects from the earlier review: the search-index-to-data-ai-config-row-id anchor contract, the hero not leaking its raw preference id, catch-all rows being disabled while AI is off, preference-ownership uniqueness/coverage across categories, and centerInBody's scroll arithmetic (including that it never touches an ancestor's scroll position). I checked the math in the centerInBody tests by hand against the implementation and it's correct, and the stub shapes used (AiSettingsRowService, getSectionRefs, getOwnedPreferenceIds) match the real APIs. No issues found in this commit.

I also re-checked the one loose end from my last review: the stale JSDoc on AiSettingRowGear claiming it opens PreferenceMenus.PREFERENCE_EDITOR_CONTEXT_MENU now correctly references AI_CONFIGURATION_SETTING_CONTEXT_MENU. That thread was already marked resolved by the maintainer, so nothing further needed there.

All previously open threads are resolved, and I found nothing new to flag in this round. No inline comments this time.