nes: support /models on proxy and model picker #2266

ulugbekna · 2025-11-28T18:00:54Z

No description provided.

I want to have `InlineEditFeature`'s code as reusable pieces such that a SingularCompletionsProvider can reuse those

Copilot

Pull request overview

This PR adds support for fetching available models from a /models endpoint on the proxy server and implements a model picker UI for inline edits. The implementation includes:

A new InlineEditsModelService that fetches and manages the list of available models from the proxy endpoint
Integration of the model service into the inline completion provider to expose model selection UI
Refactoring of XtabProvider to use the centralized model service instead of duplicated config parsing logic
A new joint completions provider that unifies completions and inline edits providers
Configuration updates to support model preferences and enable the feature via experiment flags

Reviewed changes

Copilot reviewed 16 out of 16 changed files in this pull request and generated 9 comments.

Show a summary per file

File	Description
src/platform/inlineEdits/node/inlineEditsModelService.ts	New service that fetches models from `/models` endpoint and manages model selection
src/platform/inlineEdits/common/inlineEditsModelService.ts	Service interface defining the model service contract
src/platform/inlineEdits/common/dataTypes/inlineEditsModelsTypes.ts	Wire type definitions for the /models API response
src/platform/inlineEdits/common/dataTypes/xtabPromptOptions.ts	Added type guard function for validating prompting strategy values
src/platform/configuration/common/configurationService.ts	Added new config keys for model picker and preferred model settings
src/extension/xtab/node/xtabProvider.ts	Refactored to use the new model service instead of duplicating config parsing logic
src/extension/test/vscode-node/services.ts	Registered model service in test service collection
src/extension/test/node/services.ts	Registered model service in unit test service collection
src/extension/inlineEdits/vscode-node/jointInlineCompletionProvider.ts	New unified provider that combines completions and inline edits with fallback logic
src/extension/inlineEdits/vscode-node/inlineEditProviderFeature.ts	Refactored to extract reusable logic for joint provider
src/extension/inlineEdits/vscode-node/inlineCompletionProvider.ts	Extended to support model picker UI via VS Code API
src/extension/extension/vscode-node/services.ts	Registered model service in main service collection
src/extension/extension/vscode-node/contributions.ts	Replaced separate contributions with joint provider contribution
src/extension/completions-core/vscode-node/extension/src/inlineCompletion.ts	Added undefined handlers for compatibility with joint provider
package.nls.json	Added localization string for preferred model setting
package.json	Added configuration for preferred model setting

Comments suppressed due to low confidence (1)

src/platform/inlineEdits/node/inlineEditsModelService.ts:1

The IN_TESTING flag is hardcoded to false. This should be removed or controlled via a proper configuration/environment variable to avoid accidentally leaving test code paths active in production. Consider using an experiment flag or environment variable instead.

/*---------------------------------------------------------------------------------------------

Copilot · 2025-11-28T18:05:37Z

src/platform/inlineEdits/node/inlineEditsModelService.ts

+		try {
+			r = await this._fetchService.fetch(url, {
+				headers: {
+					'Authorization': copilotToken.token,


The Authorization header is missing the 'Bearer ' prefix. Based on the codebase pattern (see src/platform/networking/common/networking.ts:314 and other examples), the header should be 'Authorization': \Bearer ${copilotToken.token}`` to follow the OAuth 2.0 Bearer token standard.

Suggested change

'Authorization': copilotToken.token,

'Authorization': `Bearer ${copilotToken.token}`,

Copilot · 2025-11-28T18:05:37Z

src/platform/inlineEdits/node/inlineEditsModelService.ts

+		let parsedConfig: ModelConfiguration | undefined;
+		try {
+			parsedConfig = JSON.parse(configString);
+			// FIXME@ulugbekna: validate parsedConfig structure


The parsed model configuration from JSON is not validated against the expected structure. Use the existing MODEL_CONFIGURATION_VALIDATOR from xtabPromptOptions.ts to validate the parsed config before using it, similar to how the non-string version is validated.

Copilot · 2025-11-28T18:05:37Z

src/platform/inlineEdits/common/dataTypes/xtabPromptOptions.ts

 }

+export function isPromptingStrategy(value: string): value is PromptingStrategy {
+	return Object.values(PromptingStrategy).includes(value as PromptingStrategy);


The type narrowing uses a cast value as PromptingStrategy before checking if it's included. While this works, it's safer to check without the cast: return (Object.values(PromptingStrategy) as string[]).includes(value); to avoid potential type confusion.

Suggested change

return Object.values(PromptingStrategy).includes(value as PromptingStrategy);

return (Object.values(PromptingStrategy) as string[]).includes(value);

Copilot · 2025-11-28T18:05:38Z

src/platform/configuration/common/configurationService.ts

 		export const InlineEditsNextCursorPredictionDisplayLine = defineAndMigrateExpSetting<boolean>('chat.advanced.inlineEdits.nextCursorPrediction.displayLine', 'chat.inlineEdits.nextCursorPrediction.displayLine', true);
 		export const InlineEditsNextCursorPredictionCurrentFileMaxTokens = defineAndMigrateExpSetting<number>('chat.advanced.inlineEdits.nextCursorPrediction.currentFileMaxTokens', 'chat.inlineEdits.nextCursorPrediction.currentFileMaxTokens', xtabPromptOptions.DEFAULT_OPTIONS.currentFile.maxTokens);
 		export const InlineEditsRenameSymbolSuggestions = defineSetting<boolean>('chat.inlineEdits.renameSymbolSuggestions', ConfigType.ExperimentBased, { defaultValue: false, teamDefaultValue: true });
+		export const InlineEditsPreferredModel = defineSetting<string | "none">('nextEditSuggestions.preferredModel', ConfigType.ExperimentBased, "none");


The configuration key nextEditSuggestions.preferredModel is inconsistent with the namespace convention. Other inline edits settings use chat.inlineEdits.* or chat.advanced.inlineEdits.* prefixes. This should be chat.inlineEdits.preferredModel or similar to match the existing pattern.

Suggested change

export const InlineEditsPreferredModel = defineSetting<string | "none">('nextEditSuggestions.preferredModel', ConfigType.ExperimentBased, "none");

export const InlineEditsPreferredModel = defineSetting<string | "none">('chat.inlineEdits.preferredModel', ConfigType.ExperimentBased, "none");

Copilot · 2025-11-28T18:05:38Z

package.json

 							"onExp"
 						]
 					},
+					"github.copilot.nextEditSuggestions.preferredModel": {


The package.json configuration key github.copilot.nextEditSuggestions.preferredModel is inconsistent with other inline edits settings. Other settings use github.copilot.chat.inlineEdits.* prefix. This should match the pattern for consistency.

Suggested change

"github.copilot.nextEditSuggestions.preferredModel": {

"github.copilot.chat.inlineEdits.nextEditSuggestions.preferredModel": {

Copilot · 2025-11-28T18:05:38Z

src/platform/inlineEdits/node/inlineEditsModelService.ts

+		try {
+			const jsonData: unknown = await r.json();
+			if (!WireTypes.ModelList.is(jsonData)) {
+				throw new Error('Invalid model list response'); // TODO@ulugbekna: add telemetry


Missing telemetry for invalid model list responses. This error case should send telemetry similar to the incorrectNesModelConfig event to track API format issues or potential attacks.

Suggested change

throw new Error('Invalid model list response'); // TODO@ulugbekna: add telemetry

/* __GDPR__

"invalidModelListResponse" : {

"owner": "ulugbekna",

"comment": "Capture if /models API returns an invalid response format.",

"errorMessage": { "classification": "SystemMetaData", "purpose": "FeatureInsight", "comment": "Error message for invalid model list response." },

"responseValue": { "classification": "SystemMetaData", "purpose": "FeatureInsight", "comment": "The invalid response received from the /models endpoint, stringified." }

}

*/

this._telemetryService.sendMSFTTelemetryEvent('invalidModelListResponse', {

errorMessage: 'Invalid model list response',

responseValue: JSON.stringify(jsonData)

});

throw new Error('Invalid model list response');

Copilot · 2025-11-28T18:05:39Z

src/platform/inlineEdits/node/inlineEditsModelService.ts

+					}
+				});
+
+				// FIXME@ulugbekna: should I manually run it once here?


This FIXME comment suggests uncertainty about the implementation. Either the manual run at lines 139-145 is necessary and the comment should explain why, or it's not necessary and should be removed. The comment should be resolved before merging.

Suggested change

// FIXME@ulugbekna: should I manually run it once here?

// Manually run once to ensure model state is initialized immediately,

// as autorun will only trigger on observable changes after this point.

Copilot · 2025-11-28T18:05:39Z

src/platform/inlineEdits/node/inlineEditsModelService.ts

+			if (preferredModelName !== 'none' && models.some(m => m.modelName === preferredModelName)) {
+				this._currentModelId = preferredModelName;
+			} else if (models.length > 0) {
+				this._currentModelId = models[0].modelName; // FIXME@ulugbekna: I think the first entry should be modelConfigurationString maybe


This FIXME suggests the model selection priority may be incorrect. The comment indicates that modelConfigurationString should take precedence over the first model in the list. This should be clarified and implemented correctly or the FIXME resolved.

Copilot · 2025-11-28T18:05:39Z

src/platform/inlineEdits/node/inlineEditsModelService.ts

+					return {
+						modelName: m.name,
+						promptingStrategy: m.capabilities.promptStrategy,
+						includeTagsInCurrentFile: false, // FIXME@ulugbekna: determine this based on model capabilities and config


The includeTagsInCurrentFile property is hardcoded to false for all fetched models. The FIXME comment indicates this should be determined from model capabilities in the response. Consider adding this capability field to WireTypes.Capabilities.t or documenting why the hardcoded value is acceptable.

Suggested change

includeTagsInCurrentFile: false, // FIXME@ulugbekna: determine this based on model capabilities and config

includeTagsInCurrentFile: !!m.capabilities.includeTagsInCurrentFile,

ulugbekna added 3 commits November 28, 2025 11:19

nes: refactor: separate InlineEditFeature from its contribution

b0bb5b3

I want to have `InlineEditFeature`'s code as reusable pieces such that a SingularCompletionsProvider can reuse those

completions-core: use a more precise return type for provide method

9dcb76c

nes: implement JointCompletionsProvider

aa3d1d3

Copilot AI review requested due to automatic review settings November 28, 2025 18:00

ulugbekna self-assigned this Nov 28, 2025

nes: support /models on proxy and model picker

be6925a

ulugbekna force-pushed the ulugbekna/slash-models branch from 98e979e to be6925a Compare November 28, 2025 18:01

Copilot started reviewing on behalf of ulugbekna November 28, 2025 18:01 View session

vs-code-engineering bot added this to the November 2025 milestone Nov 28, 2025

Copilot finished reviewing on behalf of ulugbekna November 28, 2025 18:03

Copilot AI reviewed Nov 28, 2025

View reviewed changes

lszomoru approved these changes Nov 28, 2025

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

nes: support /models on proxy and model picker #2266

nes: support /models on proxy and model picker #2266

ulugbekna commented Nov 28, 2025

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI Nov 28, 2025

Uh oh!

Copilot AI Nov 28, 2025

Uh oh!

Copilot AI Nov 28, 2025

Uh oh!

Copilot AI Nov 28, 2025

Uh oh!

Copilot AI Nov 28, 2025

Uh oh!

Copilot AI Nov 28, 2025

Uh oh!

Copilot AI Nov 28, 2025

Uh oh!

Copilot AI Nov 28, 2025

Uh oh!

Copilot AI Nov 28, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

	'Authorization': copilotToken.token,
	'Authorization': `Bearer ${copilotToken.token}`,

	return Object.values(PromptingStrategy).includes(value as PromptingStrategy);
	return (Object.values(PromptingStrategy) as string[]).includes(value);

	export const InlineEditsPreferredModel = defineSetting<string \| "none">('nextEditSuggestions.preferredModel', ConfigType.ExperimentBased, "none");
	export const InlineEditsPreferredModel = defineSetting<string \| "none">('chat.inlineEdits.preferredModel', ConfigType.ExperimentBased, "none");

	"github.copilot.nextEditSuggestions.preferredModel": {
	"github.copilot.chat.inlineEdits.nextEditSuggestions.preferredModel": {

-				throw new Error('Invalid model list response'); // TODO@ulugbekna: add telemetry
+				/* __GDPR__
+					"invalidModelListResponse" : {
+						"owner": "ulugbekna",
+						"comment": "Capture if /models API returns an invalid response format.",
+						"errorMessage": { "classification": "SystemMetaData", "purpose": "FeatureInsight", "comment": "Error message for invalid model list response." },
+						"responseValue": { "classification": "SystemMetaData", "purpose": "FeatureInsight", "comment": "The invalid response received from the /models endpoint, stringified." }
+					}
+				*/
+				this._telemetryService.sendMSFTTelemetryEvent('invalidModelListResponse', {
+					errorMessage: 'Invalid model list response',
+					responseValue: JSON.stringify(jsonData)
+				});
+				throw new Error('Invalid model list response');

	// FIXME@ulugbekna: should I manually run it once here?
	// Manually run once to ensure model state is initialized immediately,
	// as autorun will only trigger on observable changes after this point.

	includeTagsInCurrentFile: false, // FIXME@ulugbekna: determine this based on model capabilities and config
	includeTagsInCurrentFile: !!m.capabilities.includeTagsInCurrentFile,

nes: support /models on proxy and model picker #2266

Are you sure you want to change the base?

nes: support /models on proxy and model picker #2266

Conversation

ulugbekna commented Nov 28, 2025

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Nov 28, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Nov 28, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Nov 28, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Nov 28, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Nov 28, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Nov 28, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Nov 28, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Nov 28, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Nov 28, 2025

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants