From c4c7ef5c48a797640f277d0bcf67010cf0e05936 Mon Sep 17 00:00:00 2001
From: Johannes <johannes.rieken@gmail.com>
Date: Fri, 3 May 2024 15:42:29 +0200
Subject: [PATCH 1/7] API-todos for language model namespace

---
 .../vscode.proposed.languageModels.d.ts       | 27 ++++++++++++++++++-
 1 file changed, 26 insertions(+), 1 deletion(-)

diff --git a/src/vscode-dts/vscode.proposed.languageModels.d.ts b/src/vscode-dts/vscode.proposed.languageModels.d.ts
index 72db29500b9..b2d11b9ac21 100644
--- a/src/vscode-dts/vscode.proposed.languageModels.d.ts
+++ b/src/vscode-dts/vscode.proposed.languageModels.d.ts
@@ -12,14 +12,36 @@ declare module 'vscode' {
 	 */
 	export interface LanguageModelChatResponse {
 
+		//TODO@API give this some structure
+		// https://github.com/openai/openai-openapi/blob/master/openapi.yaml#L7700
+		// https://github.com/ollama/ollama/blob/main/docs/api.md#response-7
+		// https://docs.anthropic.com/claude/reference/messages_post
+
 		/**
 		 * An async iterable that is a stream of text chunks forming the overall response.
 		 *
-		 * *Note* that this stream will error when during data receiving an error occurrs.
+		 * *Note* that this stream will error when during data receiving an error occurs. Consumers of
+		 * the stream should handle the errors accordingly.
+		 *
+		 * @example
+		 * ```ts
+		 * try {
+		 *   // consume stream
+		 *   for await (const chunk of response.stream) {
+		 *    console.log(chunk);
+		 *   }
+		 *
+		 * } catch(e) {
+		 *   // stream ended with an error
+		 *   console.error(e);
+		 * }
+		 * ```
 		 */
 		stream: AsyncIterable<string>;
 	}
 
+	// TODO@API have just LanguageModelChatMessage & LanguageModelChatMessageRole
+
 	/**
 	 * A language model message that represents a system message.
 	 *
@@ -115,6 +137,7 @@ declare module 'vscode' {
 		/**
 		 * The version of the language model.
 		 */
+		// TODO@API drop this for now?
 		readonly version: string;
 
 		/**
@@ -123,6 +146,8 @@ declare module 'vscode' {
 		 *
 		 * @see {@link lm.sendChatRequest}
 		 */
+		// TODO@API is this optional?
+		// TODO@API rename: contextLength?
 		readonly tokens: number;
 	}
 

From d5b3c2f1c74ad3dd54f9f37a3a6f0e07cc73604f Mon Sep 17 00:00:00 2001
From: Johannes <johannes.rieken@gmail.com>
Date: Mon, 6 May 2024 08:38:55 +0200
Subject: [PATCH 2/7] notes

---
 .../vscode.proposed.languageModels.d.ts       | 47 ++++++++++++++++++-
 1 file changed, 46 insertions(+), 1 deletion(-)

diff --git a/src/vscode-dts/vscode.proposed.languageModels.d.ts b/src/vscode-dts/vscode.proposed.languageModels.d.ts
index b2d11b9ac21..bd2977ba5ee 100644
--- a/src/vscode-dts/vscode.proposed.languageModels.d.ts
+++ b/src/vscode-dts/vscode.proposed.languageModels.d.ts
@@ -13,7 +13,7 @@ declare module 'vscode' {
 	export interface LanguageModelChatResponse {
 
 		//TODO@API give this some structure
-		// https://github.com/openai/openai-openapi/blob/master/openapi.yaml#L7700
+		// https://github.com/openai/openai-openapi/blob/master/openapi.yaml#L7700, https://platform.openai.com/docs/guides/text-generation/chat-completions-api
 		// https://github.com/ollama/ollama/blob/main/docs/api.md#response-7
 		// https://docs.anthropic.com/claude/reference/messages_post
 
@@ -40,7 +40,48 @@ declare module 'vscode' {
 		stream: AsyncIterable<string>;
 	}
 
+	export enum LanguageModelChatMessageRole {
+		User,
+		Assistant,
+		// System see https://github.com/anthropics/anthropic-sdk-typescript/blob/c2da9604646ff103fbdbca016a9a9d49b03b387b/src/resources/messages.ts#L384
+	}
+
+	export class LanguageModelChatMessage2 {
+		/**
+		 * The role of this message.
+		 */
+		role: LanguageModelChatMessageRole;
+
+		/**
+		 * The content of this message.
+		 */
+		content: string;
+
+		/**
+		 * The optional name of a user for this message.
+		 */
+		name: string | undefined;
+
+		/**
+		 * Create a new user message.
+		 *
+		 * @param content The content of the message.
+		 * @param name The optional name of a user for the message.
+		 */
+		constructor(role: LanguageModelChatMessageRole, content: string, name?: string);
+	}
+
+	export interface LanguageModelChatResponse2 {
+
+		message: {
+			role: LanguageModelChatMessageRole;
+			content: AsyncIterable<string>;
+		};
+	}
+
 	// TODO@API have just LanguageModelChatMessage & LanguageModelChatMessageRole
+	// TODO@API don't have SystemChatMessage because it encourages to mix system and user messages
+	// but AFAIK system message should/must come first (https://github.com/anthropics/anthropic-sdk-typescript/blob/c2da9604646ff103fbdbca016a9a9d49b03b387b/src/resources/messages.ts#L384)
 
 	/**
 	 * A language model message that represents a system message.
@@ -209,6 +250,8 @@ declare module 'vscode' {
 	 */
 	export interface LanguageModelChatRequestOptions {
 
+		// TODO@API add short name for the feature using the language model
+
 		/**
 		 * A human-readable message that explains why access to a language model is needed and what feature is enabled by it.
 		 */
@@ -220,6 +263,8 @@ declare module 'vscode' {
 		// TODO@API Revisit this, how do you do the first request?
 		silent?: boolean;
 
+		// TODO@API add system messages here?
+
 		/**
 		 * A set of options that control the behavior of the language model. These options are specific to the language model
 		 * and need to be lookup in the respective documentation.

From 22efbdf96cca32d1bad2fac39400a750c20027eb Mon Sep 17 00:00:00 2001
From: Johannes <johannes.rieken@gmail.com>
Date: Mon, 6 May 2024 11:00:17 +0200
Subject: [PATCH 3/7] * return `LanguageModelChatMessageRole` and the simple
 chat message `LanguageModelChatMessage2` because all types became same, same
 * deprecate the other types and prepare deletion * move System-role into its
 own proposal because of API and RAI concerns * more API notes

---
 .../workbench/api/common/extHost.api.impl.ts  | 13 ++--
 .../api/common/extHostLanguageModels.ts       | 43 ++++++++---
 .../api/common/extHostTypeConverters.ts       | 23 ++++++
 src/vs/workbench/api/common/extHostTypes.ts   | 19 +++++
 .../common/extensionsApiProposals.ts          |  1 +
 .../vscode.proposed.chatProvider.d.ts         |  2 +
 .../vscode.proposed.languageModelSystem.d.ts  | 17 +++++
 .../vscode.proposed.languageModels.d.ts       | 71 ++++++++++---------
 8 files changed, 140 insertions(+), 49 deletions(-)
 create mode 100644 src/vscode-dts/vscode.proposed.languageModelSystem.d.ts

diff --git a/src/vs/workbench/api/common/extHost.api.impl.ts b/src/vs/workbench/api/common/extHost.api.impl.ts
index 3caf8e8556a..23ef3d71187 100644
--- a/src/vs/workbench/api/common/extHost.api.impl.ts
+++ b/src/vs/workbench/api/common/extHost.api.impl.ts
@@ -1438,7 +1438,7 @@ export function createApiFactoryAndRegisterActors(accessor: ServicesAccessor): I
 				checkProposedApiEnabled(extension, 'languageModels');
 				return extHostLanguageModels.onDidChangeProviders(listener, thisArgs, disposables);
 			},
-			sendChatRequest(languageModel: string, messages: vscode.LanguageModelChatMessage[], options: vscode.LanguageModelChatRequestOptions, token: vscode.CancellationToken) {
+			sendChatRequest(languageModel: string, messages: (vscode.LanguageModelChatMessage | vscode.LanguageModelChatMessage2)[], options: vscode.LanguageModelChatRequestOptions, token: vscode.CancellationToken) {
 				checkProposedApiEnabled(extension, 'languageModels');
 				return extHostLanguageModels.sendChatRequest(extension, languageModel, messages, options, token);
 			},
@@ -1707,12 +1707,11 @@ export function createApiFactoryAndRegisterActors(accessor: ServicesAccessor): I
 			ChatRequestTurn: extHostTypes.ChatRequestTurn,
 			ChatResponseTurn: extHostTypes.ChatResponseTurn,
 			ChatLocation: extHostTypes.ChatLocation,
-			LanguageModelChatSystemMessage: extHostTypes.LanguageModelChatSystemMessage,
-			LanguageModelChatUserMessage: extHostTypes.LanguageModelChatUserMessage,
-			LanguageModelChatAssistantMessage: extHostTypes.LanguageModelChatAssistantMessage,
-			LanguageModelSystemMessage: extHostTypes.LanguageModelChatSystemMessage, // TODO@jrieken REMOVE
-			LanguageModelUserMessage: extHostTypes.LanguageModelChatUserMessage, // TODO@jrieken REMOVE
-			LanguageModelAssistantMessage: extHostTypes.LanguageModelChatAssistantMessage, // TODO@jrieken REMOVE
+			LanguageModelChatMessageRole: extHostTypes.LanguageModelChatMessageRole,
+			LanguageModelChatMessage2: extHostTypes.LanguageModelChatMessage2,
+			LanguageModelChatSystemMessage: extHostTypes.LanguageModelChatSystemMessage,// TODO@jrieken REMOVE
+			LanguageModelChatUserMessage: extHostTypes.LanguageModelChatUserMessage,// TODO@jrieken REMOVE
+			LanguageModelChatAssistantMessage: extHostTypes.LanguageModelChatAssistantMessage,// TODO@jrieken REMOVE
 			LanguageModelError: extHostTypes.LanguageModelError,
 			NewSymbolName: extHostTypes.NewSymbolName,
 			NewSymbolNameTag: extHostTypes.NewSymbolNameTag,
diff --git a/src/vs/workbench/api/common/extHostLanguageModels.ts b/src/vs/workbench/api/common/extHostLanguageModels.ts
index d0202a84451..4a707110621 100644
--- a/src/vs/workbench/api/common/extHostLanguageModels.ts
+++ b/src/vs/workbench/api/common/extHostLanguageModels.ts
@@ -7,7 +7,7 @@ import { CancellationToken } from 'vs/base/common/cancellation';
 import { IDisposable, toDisposable } from 'vs/base/common/lifecycle';
 import { ExtHostLanguageModelsShape, MainContext, MainThreadLanguageModelsShape } from 'vs/workbench/api/common/extHost.protocol';
 import * as typeConvert from 'vs/workbench/api/common/extHostTypeConverters';
-import { LanguageModelError } from 'vs/workbench/api/common/extHostTypes';
+import * as extHostTypes from 'vs/workbench/api/common/extHostTypes';
 import type * as vscode from 'vscode';
 import { Progress } from 'vs/platform/progress/common/progress';
 import { IChatMessage, IChatResponseFragment, ILanguageModelChatMetadata } from 'vs/workbench/contrib/chat/common/languageModels';
@@ -22,6 +22,7 @@ import { IExtHostRpcService } from 'vs/workbench/api/common/extHostRpcService';
 import { IExtHostAuthentication } from 'vs/workbench/api/common/extHostAuthentication';
 import { ILogService } from 'vs/platform/log/common/log';
 import { Iterable } from 'vs/base/common/iterator';
+import { checkProposedApiEnabled } from 'vs/workbench/services/extensions/common/extensions';
 
 export interface IExtHostLanguageModels extends ExtHostLanguageModels { }
 
@@ -185,7 +186,11 @@ export class ExtHostLanguageModels implements ExtHostLanguageModelsShape {
 			this._proxy.$handleProgressChunk(requestId, { index: fragment.index, part: fragment.part });
 		});
 
-		return data.provider.provideLanguageModelResponse2(messages.map(typeConvert.LanguageModelMessage.to), options, ExtensionIdentifier.toKey(from), progress, token);
+		if (data.provider.provideLanguageModelResponse) {
+			return data.provider.provideLanguageModelResponse(messages.map(typeConvert.LanguageModelChatMessage.to), options, ExtensionIdentifier.toKey(from), progress, token);
+		} else {
+			return data.provider.provideLanguageModelResponse2(messages.map(typeConvert.LanguageModelMessage.to), options, ExtensionIdentifier.toKey(from), progress, token);
+		}
 	}
 
 
@@ -259,25 +264,27 @@ export class ExtHostLanguageModels implements ExtHostLanguageModelsShape {
 		}
 	}
 
-	async sendChatRequest(extension: IExtensionDescription, languageModelId: string, messages: vscode.LanguageModelChatMessage[], options: vscode.LanguageModelChatRequestOptions, token: CancellationToken) {
+	async sendChatRequest(extension: IExtensionDescription, languageModelId: string, messages: (vscode.LanguageModelChatMessage | vscode.LanguageModelChatMessage2)[], options: vscode.LanguageModelChatRequestOptions, token: CancellationToken) {
+
+		const internalMessages: IChatMessage[] = this._convertMessages(extension, messages);
 
 		const from = extension.identifier;
 		const metadata = await this._proxy.$prepareChatAccess(from, languageModelId, options.justification);
 
 		if (!metadata || !this._allLanguageModelData.has(languageModelId)) {
-			throw LanguageModelError.NotFound(`Language model '${languageModelId}' is unknown.`);
+			throw extHostTypes.LanguageModelError.NotFound(`Language model '${languageModelId}' is unknown.`);
 		}
 
 		if (this._isUsingAuth(from, metadata)) {
 			const success = await this._getAuthAccess(extension, { identifier: metadata.extension, displayName: metadata.auth.providerLabel }, options.justification, options.silent);
 
 			if (!success || !this._modelAccessList.get(from)?.has(metadata.extension)) {
-				throw LanguageModelError.NoPermissions(`Language model '${languageModelId}' cannot be used by '${from.value}'.`);
+				throw extHostTypes.LanguageModelError.NoPermissions(`Language model '${languageModelId}' cannot be used by '${from.value}'.`);
 			}
 		}
 
 		const requestId = (Math.random() * 1e6) | 0;
-		const requestPromise = this._proxy.$fetchResponse(from, languageModelId, requestId, messages.map(typeConvert.LanguageModelMessage.from), options.modelOptions ?? {}, token);
+		const requestPromise = this._proxy.$fetchResponse(from, languageModelId, requestId, internalMessages, options.modelOptions ?? {}, token);
 
 		const barrier = new Barrier();
 
@@ -303,11 +310,11 @@ export class ExtHostLanguageModels implements ExtHostLanguageModelsShape {
 		await barrier.wait();
 
 		if (error) {
-			if (error.name === LanguageModelError.name) {
+			if (error.name === extHostTypes.LanguageModelError.name) {
 				throw error;
 			}
 
-			throw new LanguageModelError(
+			throw new extHostTypes.LanguageModelError(
 				`Language model '${languageModelId}' errored, check cause for more details`,
 				'Unknown',
 				error
@@ -317,6 +324,24 @@ export class ExtHostLanguageModels implements ExtHostLanguageModelsShape {
 		return res.apiObject;
 	}
 
+	private _convertMessages(extension: IExtensionDescription, messages: (vscode.LanguageModelChatMessage2 | vscode.LanguageModelChatMessage)[]) {
+		const internalMessages: IChatMessage[] = [];
+		for (const message of messages) {
+			if (message instanceof extHostTypes.LanguageModelChatMessage2) {
+				if (message.role as number === extHostTypes.LanguageModelChatMessageRole.System) {
+					checkProposedApiEnabled(extension, 'languageModelSystem');
+				}
+				internalMessages.push(typeConvert.LanguageModelChatMessage.from(message));
+			} else {
+				if (message instanceof extHostTypes.LanguageModelChatSystemMessage) {
+					checkProposedApiEnabled(extension, 'languageModelSystem');
+				}
+				internalMessages.push(typeConvert.LanguageModelMessage.from(message));
+			}
+		}
+		return internalMessages;
+	}
+
 	async $handleResponseFragment(requestId: number, chunk: IChatResponseFragment): Promise<void> {
 		const data = this._pendingRequest.get(requestId);//.report(chunk);
 		if (data) {
@@ -376,7 +401,7 @@ export class ExtHostLanguageModels implements ExtHostLanguageModelsShape {
 
 		const data = this._allLanguageModelData.get(languageModelId);
 		if (!data) {
-			throw LanguageModelError.NotFound(`Language model '${languageModelId}' is unknown.`);
+			throw extHostTypes.LanguageModelError.NotFound(`Language model '${languageModelId}' is unknown.`);
 		}
 
 		const local = Iterable.find(this._languageModels.values(), candidate => candidate.languageModelId === languageModelId);
diff --git a/src/vs/workbench/api/common/extHostTypeConverters.ts b/src/vs/workbench/api/common/extHostTypeConverters.ts
index d163386bf00..f9571a57858 100644
--- a/src/vs/workbench/api/common/extHostTypeConverters.ts
+++ b/src/vs/workbench/api/common/extHostTypeConverters.ts
@@ -2239,6 +2239,29 @@ export namespace ChatFollowup {
 	}
 }
 
+
+export namespace LanguageModelChatMessage {
+
+	export function to(message: chatProvider.IChatMessage): vscode.LanguageModelChatMessage2 {
+		switch (message.role) {
+			case chatProvider.ChatMessageRole.System: return new types.LanguageModelChatMessage2(<number>types.LanguageModelChatMessageRole.System, message.content);
+			case chatProvider.ChatMessageRole.User: return new types.LanguageModelChatMessage2(<number>types.LanguageModelChatMessageRole.User, message.content);
+			case chatProvider.ChatMessageRole.Assistant: return new types.LanguageModelChatMessage2(<number>types.LanguageModelChatMessageRole.Assistant, message.content);
+		}
+	}
+
+	export function from(message: vscode.LanguageModelChatMessage2): chatProvider.IChatMessage {
+		switch (message.role as types.LanguageModelChatMessageRole) {
+			case types.LanguageModelChatMessageRole.System: return { role: chatProvider.ChatMessageRole.System, content: message.content };
+			case types.LanguageModelChatMessageRole.User: return { role: chatProvider.ChatMessageRole.User, content: message.content };
+			case types.LanguageModelChatMessageRole.Assistant: return { role: chatProvider.ChatMessageRole.Assistant, content: message.content };
+		}
+	}
+}
+
+/**
+ * @deprecated
+ */
 export namespace LanguageModelMessage {
 
 	export function to(message: chatProvider.IChatMessage): vscode.LanguageModelChatMessage {
diff --git a/src/vs/workbench/api/common/extHostTypes.ts b/src/vs/workbench/api/common/extHostTypes.ts
index 9a85e2f08e8..c52e1c26cc2 100644
--- a/src/vs/workbench/api/common/extHostTypes.ts
+++ b/src/vs/workbench/api/common/extHostTypes.ts
@@ -4424,6 +4424,25 @@ export enum ChatLocation {
 	Editor = 4,
 }
 
+export enum LanguageModelChatMessageRole {
+	User = 1,
+	Assistant = 2,
+	System = 3
+}
+
+export class LanguageModelChatMessage2 implements vscode.LanguageModelChatMessage2 {
+
+	role: vscode.LanguageModelChatMessageRole;
+	content: string;
+	name: string | undefined;
+
+	constructor(role: vscode.LanguageModelChatMessageRole, content: string, name?: string) {
+		this.role = role;
+		this.content = content;
+		this.name = name;
+	}
+}
+
 export class LanguageModelChatSystemMessage {
 	content: string;
 
diff --git a/src/vs/workbench/services/extensions/common/extensionsApiProposals.ts b/src/vs/workbench/services/extensions/common/extensionsApiProposals.ts
index 402560ad1bd..c8c0a5d0e74 100644
--- a/src/vs/workbench/services/extensions/common/extensionsApiProposals.ts
+++ b/src/vs/workbench/services/extensions/common/extensionsApiProposals.ts
@@ -74,6 +74,7 @@ export const allApiProposals = Object.freeze({
 	interactive: 'https://raw.githubusercontent.com/microsoft/vscode/main/src/vscode-dts/vscode.proposed.interactive.d.ts',
 	interactiveWindow: 'https://raw.githubusercontent.com/microsoft/vscode/main/src/vscode-dts/vscode.proposed.interactiveWindow.d.ts',
 	ipc: 'https://raw.githubusercontent.com/microsoft/vscode/main/src/vscode-dts/vscode.proposed.ipc.d.ts',
+	languageModelSystem: 'https://raw.githubusercontent.com/microsoft/vscode/main/src/vscode-dts/vscode.proposed.languageModelSystem.d.ts',
 	languageModels: 'https://raw.githubusercontent.com/microsoft/vscode/main/src/vscode-dts/vscode.proposed.languageModels.d.ts',
 	languageStatusText: 'https://raw.githubusercontent.com/microsoft/vscode/main/src/vscode-dts/vscode.proposed.languageStatusText.d.ts',
 	mappedEditsProvider: 'https://raw.githubusercontent.com/microsoft/vscode/main/src/vscode-dts/vscode.proposed.mappedEditsProvider.d.ts',
diff --git a/src/vscode-dts/vscode.proposed.chatProvider.d.ts b/src/vscode-dts/vscode.proposed.chatProvider.d.ts
index 504027744c4..6651c4ff859 100644
--- a/src/vscode-dts/vscode.proposed.chatProvider.d.ts
+++ b/src/vscode-dts/vscode.proposed.chatProvider.d.ts
@@ -19,6 +19,8 @@ declare module 'vscode' {
 
 		onDidReceiveLanguageModelResponse2?: Event<{ readonly extensionId: string; readonly participant?: string; readonly tokenCount?: number }>;
 
+		provideLanguageModelResponse?(messages: LanguageModelChatMessage2[], options: { [name: string]: any }, extensionId: string, progress: Progress<ChatResponseFragment>, token: CancellationToken): Thenable<any>;
+
 		provideLanguageModelResponse2(messages: LanguageModelChatMessage[], options: { [name: string]: any }, extensionId: string, progress: Progress<ChatResponseFragment>, token: CancellationToken): Thenable<any>;
 
 		provideTokenCount(text: string | LanguageModelChatMessage, token: CancellationToken): Thenable<number>;
diff --git a/src/vscode-dts/vscode.proposed.languageModelSystem.d.ts b/src/vscode-dts/vscode.proposed.languageModelSystem.d.ts
new file mode 100644
index 00000000000..54e1da27a5b
--- /dev/null
+++ b/src/vscode-dts/vscode.proposed.languageModelSystem.d.ts
@@ -0,0 +1,17 @@
+/*---------------------------------------------------------------------------------------------
+ *  Copyright (c) Microsoft Corporation. All rights reserved.
+ *  Licensed under the MIT License. See License.txt in the project root for license information.
+ *--------------------------------------------------------------------------------------------*/
+
+declare module 'vscode' {
+
+	// https://github.com/microsoft/vscode/issues/206265
+
+	// TODO@API don't have this dedicated type but as property, e.g anthropic doesn't have a system-role, see
+	// https://github.com/anthropics/anthropic-sdk-typescript/blob/c2da9604646ff103fbdbca016a9a9d49b03b387b/src/resources/messages.ts#L384
+	// So, we could have `LanguageModelChatRequestOptions#system` which would be more limiting but also more natural?
+
+	export enum LanguageModelChatMessageRole {
+		System = 3
+	}
+}
diff --git a/src/vscode-dts/vscode.proposed.languageModels.d.ts b/src/vscode-dts/vscode.proposed.languageModels.d.ts
index bd2977ba5ee..db75bbeb4ea 100644
--- a/src/vscode-dts/vscode.proposed.languageModels.d.ts
+++ b/src/vscode-dts/vscode.proposed.languageModels.d.ts
@@ -5,6 +5,8 @@
 
 declare module 'vscode' {
 
+	// https://github.com/microsoft/vscode/issues/206265
+
 	/**
 	 * Represents a language model response.
 	 *
@@ -12,11 +14,6 @@ declare module 'vscode' {
 	 */
 	export interface LanguageModelChatResponse {
 
-		//TODO@API give this some structure
-		// https://github.com/openai/openai-openapi/blob/master/openapi.yaml#L7700, https://platform.openai.com/docs/guides/text-generation/chat-completions-api
-		// https://github.com/ollama/ollama/blob/main/docs/api.md#response-7
-		// https://docs.anthropic.com/claude/reference/messages_post
-
 		/**
 		 * An async iterable that is a stream of text chunks forming the overall response.
 		 *
@@ -40,12 +37,35 @@ declare module 'vscode' {
 		stream: AsyncIterable<string>;
 	}
 
-	export enum LanguageModelChatMessageRole {
-		User,
-		Assistant,
-		// System see https://github.com/anthropics/anthropic-sdk-typescript/blob/c2da9604646ff103fbdbca016a9a9d49b03b387b/src/resources/messages.ts#L384
+	//TODO@API give this some structure
+	// https://github.com/openai/openai-openapi/blob/master/openapi.yaml#L7700, https://platform.openai.com/docs/guides/text-generation/chat-completions-api
+	// https://github.com/ollama/ollama/blob/main/docs/api.md#response-7
+	// https://docs.anthropic.com/claude/reference/messages_post
+	export interface LanguageModelChatResponse2 {
+
+		message: {
+			role: LanguageModelChatMessageRole;
+			content: AsyncIterable<string>;
+		};
 	}
 
+	/**
+	 * Represents the role of a chat message. This is either the user or the assistant/model.
+	 */
+	export enum LanguageModelChatMessageRole {
+		/**
+		 * The user role, e.g the human interacting with a language model.
+		 */
+		User = 1,
+
+		/**
+		 * The assistant role, e.g. the language model generating responses.
+		 */
+		// TODO@API name: model?
+		Assistant = 2
+	}
+
+	// TODO@API name: LanguageModelChatMessage once the deprecated stuff is removed
 	export class LanguageModelChatMessage2 {
 		/**
 		 * The role of this message.
@@ -71,26 +91,9 @@ declare module 'vscode' {
 		constructor(role: LanguageModelChatMessageRole, content: string, name?: string);
 	}
 
-	export interface LanguageModelChatResponse2 {
-
-		message: {
-			role: LanguageModelChatMessageRole;
-			content: AsyncIterable<string>;
-		};
-	}
-
-	// TODO@API have just LanguageModelChatMessage & LanguageModelChatMessageRole
-	// TODO@API don't have SystemChatMessage because it encourages to mix system and user messages
-	// but AFAIK system message should/must come first (https://github.com/anthropics/anthropic-sdk-typescript/blob/c2da9604646ff103fbdbca016a9a9d49b03b387b/src/resources/messages.ts#L384)
 
 	/**
-	 * A language model message that represents a system message.
-	 *
-	 * System messages provide instructions to the language model that define the context in
-	 * which user messages are interpreted.
-	 *
-	 * *Note* that a language model may choose to add additional system messages to the ones
-	 * provided by extensions.
+	 * @deprecated
 	 */
 	export class LanguageModelChatSystemMessage {
 
@@ -108,7 +111,7 @@ declare module 'vscode' {
 	}
 
 	/**
-	 * A language model message that represents a user message.
+	 * @deprecated
 	 */
 	export class LanguageModelChatUserMessage {
 
@@ -132,8 +135,7 @@ declare module 'vscode' {
 	}
 
 	/**
-	 * A language model message that represents an assistant message, usually in response to a user message
-	 * or as a sample response/reply-pair.
+	 * @deprecated
 	 */
 	export class LanguageModelChatAssistantMessage {
 
@@ -158,6 +160,7 @@ declare module 'vscode' {
 
 	/**
 	 * Different types of language model messages.
+	 * @deprecated
 	 */
 	export type LanguageModelChatMessage = LanguageModelChatSystemMessage | LanguageModelChatUserMessage | LanguageModelChatAssistantMessage;
 
@@ -263,7 +266,6 @@ declare module 'vscode' {
 		// TODO@API Revisit this, how do you do the first request?
 		silent?: boolean;
 
-		// TODO@API add system messages here?
 
 		/**
 		 * A set of options that control the behavior of the language model. These options are specific to the language model
@@ -317,7 +319,7 @@ declare module 'vscode' {
 		 * @param token A cancellation token which controls the request. See {@link CancellationTokenSource} for how to create one.
 		 * @returns A thenable that resolves to a {@link LanguageModelChatResponse}. The promise will reject when the request couldn't be made.
 		 */
-		export function sendChatRequest(languageModel: string, messages: LanguageModelChatMessage[], options: LanguageModelChatRequestOptions, token: CancellationToken): Thenable<LanguageModelChatResponse>;
+		export function sendChatRequest(languageModel: string, messages: (LanguageModelChatMessage | LanguageModelChatMessage2)[], options: LanguageModelChatRequestOptions, token: CancellationToken): Thenable<LanguageModelChatResponse>;
 
 		/**
 		 * Uses the language model specific tokenzier and computes the length in token of a given message.
@@ -329,7 +331,10 @@ declare module 'vscode' {
 		 * @param token Optional cancellation token.
 		 * @returns A thenable that resolves to the length of the message in tokens.
 		 */
-		export function computeTokenLength(languageModel: string, text: string | LanguageModelChatMessage, token?: CancellationToken): Thenable<number>;
+		// TODO@API optional?
+		// ollama has nothing
+		// anthropic suggests to count after the fact https://github.com/anthropics/anthropic-tokenizer-typescript?tab=readme-ov-file#anthropic-typescript-tokenizer
+		export function computeTokenLength(languageModel: string, text: string | LanguageModelChatMessage | LanguageModelChatMessage2, token?: CancellationToken): Thenable<number>;
 	}
 
 	/**

From fff5d2408b288504410354232bbc21f0dbda642c Mon Sep 17 00:00:00 2001
From: Johannes <johannes.rieken@gmail.com>
Date: Mon, 6 May 2024 11:16:29 +0200
Subject: [PATCH 4/7] remove `LanguageModelChatRequestOptions.silent` because
 there is `LanguageModelAccessInformation#canSendRequest`

---
 .../api/common/extHostLanguageModels.ts         |  2 +-
 .../vscode.proposed.languageModels.d.ts         | 17 +++++++----------
 2 files changed, 8 insertions(+), 11 deletions(-)

diff --git a/src/vs/workbench/api/common/extHostLanguageModels.ts b/src/vs/workbench/api/common/extHostLanguageModels.ts
index 4a707110621..89534da4033 100644
--- a/src/vs/workbench/api/common/extHostLanguageModels.ts
+++ b/src/vs/workbench/api/common/extHostLanguageModels.ts
@@ -276,7 +276,7 @@ export class ExtHostLanguageModels implements ExtHostLanguageModelsShape {
 		}
 
 		if (this._isUsingAuth(from, metadata)) {
-			const success = await this._getAuthAccess(extension, { identifier: metadata.extension, displayName: metadata.auth.providerLabel }, options.justification, options.silent);
+			const success = await this._getAuthAccess(extension, { identifier: metadata.extension, displayName: metadata.auth.providerLabel }, options.justification, false);
 
 			if (!success || !this._modelAccessList.get(from)?.has(metadata.extension)) {
 				throw extHostTypes.LanguageModelError.NoPermissions(`Language model '${languageModelId}' cannot be used by '${from.value}'.`);
diff --git a/src/vscode-dts/vscode.proposed.languageModels.d.ts b/src/vscode-dts/vscode.proposed.languageModels.d.ts
index db75bbeb4ea..8e56ee576f5 100644
--- a/src/vscode-dts/vscode.proposed.languageModels.d.ts
+++ b/src/vscode-dts/vscode.proposed.languageModels.d.ts
@@ -260,13 +260,6 @@ declare module 'vscode' {
 		 */
 		justification?: string;
 
-		/**
-		 * Do not show the consent UI if the user has not yet granted access to the language model but fail the request instead.
-		 */
-		// TODO@API Revisit this, how do you do the first request?
-		silent?: boolean;
-
-
 		/**
 		 * A set of options that control the behavior of the language model. These options are specific to the language model
 		 * and need to be lookup in the respective documentation.
@@ -300,18 +293,22 @@ declare module 'vscode' {
 		/**
 		 * Make a chat request using a language model.
 		 *
-		 * - *Note 1:* language model use may be subject to access restrictions and user consent.
+		 * - *Note 1:* language model use may be subject to access restrictions and user consent. Calling this function
+		 * for the first time (for a extension) will show a consent dialog to the user and because of that this function
+		 * must _only be called in response to a user action!_ Extension can use {@link LanguageModelAccessInformation.canSendRequest}
+		 * to check if they have the necessary permissions to make a request.
 		 *
 		 * - *Note 2:* language models are contributed by other extensions and as they evolve and change,
 		 * the set of available language models may change over time. Therefore it is strongly recommend to check
-		 * {@link languageModels} for aviailable values and handle missing language models gracefully.
+		 * {@link languageModels} for available values and handle missing language models gracefully.
 		 *
 		 * This function will return a rejected promise if making a request to the language model is not
 		 * possible. Reasons for this can be:
 		 *
 		 * - user consent not given, see {@link LanguageModelError.NoPermissions `NoPermissions`}
 		 * - model does not exist, see {@link LanguageModelError.NotFound `NotFound`}
-		 * - quota limits exceeded, see {@link LanguageModelError.cause `LanguageModelError.cause`}
+		 * - quota limits exceeded, see {@link LanguageModelError.Blocked `Blocked`}
+		 * - other issues in which case extension must check {@link LanguageModelError.cause `LanguageModelError.cause`}
 		 *
 		 * @param languageModel A language model identifier.
 		 * @param messages An array of message instances.

From 1ca4972d82badaed01c0bfa14c68943176fde9fe Mon Sep 17 00:00:00 2001
From: Johannes <johannes.rieken@gmail.com>
Date: Mon, 6 May 2024 11:21:00 +0200
Subject: [PATCH 5/7] make token and options optional when calling
 `sendChatRequest`

---
 src/vs/workbench/api/common/extHost.api.impl.ts    | 4 +++-
 src/vscode-dts/vscode.proposed.languageModels.d.ts | 2 +-
 2 files changed, 4 insertions(+), 2 deletions(-)

diff --git a/src/vs/workbench/api/common/extHost.api.impl.ts b/src/vs/workbench/api/common/extHost.api.impl.ts
index 23ef3d71187..d3c6638851a 100644
--- a/src/vs/workbench/api/common/extHost.api.impl.ts
+++ b/src/vs/workbench/api/common/extHost.api.impl.ts
@@ -1438,8 +1438,10 @@ export function createApiFactoryAndRegisterActors(accessor: ServicesAccessor): I
 				checkProposedApiEnabled(extension, 'languageModels');
 				return extHostLanguageModels.onDidChangeProviders(listener, thisArgs, disposables);
 			},
-			sendChatRequest(languageModel: string, messages: (vscode.LanguageModelChatMessage | vscode.LanguageModelChatMessage2)[], options: vscode.LanguageModelChatRequestOptions, token: vscode.CancellationToken) {
+			sendChatRequest(languageModel: string, messages: (vscode.LanguageModelChatMessage | vscode.LanguageModelChatMessage2)[], options?: vscode.LanguageModelChatRequestOptions, token?: vscode.CancellationToken) {
 				checkProposedApiEnabled(extension, 'languageModels');
+				token ??= CancellationToken.None;
+				options ??= {};
 				return extHostLanguageModels.sendChatRequest(extension, languageModel, messages, options, token);
 			},
 			computeTokenLength(languageModel: string, text: string | vscode.LanguageModelChatMessage, token?: vscode.CancellationToken) {
diff --git a/src/vscode-dts/vscode.proposed.languageModels.d.ts b/src/vscode-dts/vscode.proposed.languageModels.d.ts
index 8e56ee576f5..a54e68a5106 100644
--- a/src/vscode-dts/vscode.proposed.languageModels.d.ts
+++ b/src/vscode-dts/vscode.proposed.languageModels.d.ts
@@ -316,7 +316,7 @@ declare module 'vscode' {
 		 * @param token A cancellation token which controls the request. See {@link CancellationTokenSource} for how to create one.
 		 * @returns A thenable that resolves to a {@link LanguageModelChatResponse}. The promise will reject when the request couldn't be made.
 		 */
-		export function sendChatRequest(languageModel: string, messages: (LanguageModelChatMessage | LanguageModelChatMessage2)[], options: LanguageModelChatRequestOptions, token: CancellationToken): Thenable<LanguageModelChatResponse>;
+		export function sendChatRequest(languageModel: string, messages: (LanguageModelChatMessage | LanguageModelChatMessage2)[], options?: LanguageModelChatRequestOptions, token?: CancellationToken): Thenable<LanguageModelChatResponse>;
 
 		/**
 		 * Uses the language model specific tokenzier and computes the length in token of a given message.

From 0cfd5a3baf5f3b6d323933a440bcafbf728a65d6 Mon Sep 17 00:00:00 2001
From: Johannes <johannes.rieken@gmail.com>
Date: Mon, 6 May 2024 15:49:00 +0200
Subject: [PATCH 6/7] todo updates

---
 src/vscode-dts/vscode.proposed.languageModels.d.ts | 11 +++++------
 1 file changed, 5 insertions(+), 6 deletions(-)

diff --git a/src/vscode-dts/vscode.proposed.languageModels.d.ts b/src/vscode-dts/vscode.proposed.languageModels.d.ts
index a54e68a5106..34299a308c0 100644
--- a/src/vscode-dts/vscode.proposed.languageModels.d.ts
+++ b/src/vscode-dts/vscode.proposed.languageModels.d.ts
@@ -44,9 +44,11 @@ declare module 'vscode' {
 	export interface LanguageModelChatResponse2 {
 
 		message: {
-			role: LanguageModelChatMessageRole;
+			role: LanguageModelChatMessageRole.Assistant;
 			content: AsyncIterable<string>;
 		};
+
+
 	}
 
 	/**
@@ -61,7 +63,6 @@ declare module 'vscode' {
 		/**
 		 * The assistant role, e.g. the language model generating responses.
 		 */
-		// TODO@API name: model?
 		Assistant = 2
 	}
 
@@ -253,8 +254,6 @@ declare module 'vscode' {
 	 */
 	export interface LanguageModelChatRequestOptions {
 
-		// TODO@API add short name for the feature using the language model
-
 		/**
 		 * A human-readable message that explains why access to a language model is needed and what feature is enabled by it.
 		 */
@@ -328,10 +327,10 @@ declare module 'vscode' {
 		 * @param token Optional cancellation token.
 		 * @returns A thenable that resolves to the length of the message in tokens.
 		 */
-		// TODO@API optional?
+		// TODO@API `undefined` when the language model does not support computing token length
 		// ollama has nothing
 		// anthropic suggests to count after the fact https://github.com/anthropics/anthropic-tokenizer-typescript?tab=readme-ov-file#anthropic-typescript-tokenizer
-		export function computeTokenLength(languageModel: string, text: string | LanguageModelChatMessage | LanguageModelChatMessage2, token?: CancellationToken): Thenable<number>;
+		export function computeTokenLength(languageModel: string, text: string | LanguageModelChatMessage | LanguageModelChatMessage2, token?: CancellationToken): Thenable<number | undefined>;
 	}
 
 	/**

From 837430d8ad8dc7298dcd1ccce67f995f22182596 Mon Sep 17 00:00:00 2001
From: Johannes <johannes.rieken@gmail.com>
Date: Mon, 6 May 2024 16:27:32 +0200
Subject: [PATCH 7/7] use `LanguageModelInformation#contextLength`

---
 src/vs/workbench/api/common/extHostLanguageModels.ts | 6 +++---
 src/vscode-dts/vscode.proposed.languageModels.d.ts   | 6 +++---
 2 files changed, 6 insertions(+), 6 deletions(-)

diff --git a/src/vs/workbench/api/common/extHostLanguageModels.ts b/src/vs/workbench/api/common/extHostLanguageModels.ts
index 89534da4033..a7823af0e4c 100644
--- a/src/vs/workbench/api/common/extHostLanguageModels.ts
+++ b/src/vs/workbench/api/common/extHostLanguageModels.ts
@@ -419,12 +419,12 @@ export class ExtHostLanguageModels implements ExtHostLanguageModelsShape {
 			return undefined;
 		}
 
-		return {
+		return Object.freeze({
 			id: data.identifier,
 			name: data.name,
 			version: data.version,
-			tokens: data.tokens,
-		};
+			contextLength: data.tokens,
+		});
 	}
 
 	private readonly _languageAccessInformationExtensions = new Set<Readonly<IExtensionDescription>>();
diff --git a/src/vscode-dts/vscode.proposed.languageModels.d.ts b/src/vscode-dts/vscode.proposed.languageModels.d.ts
index 34299a308c0..be546fe75b1 100644
--- a/src/vscode-dts/vscode.proposed.languageModels.d.ts
+++ b/src/vscode-dts/vscode.proposed.languageModels.d.ts
@@ -189,11 +189,11 @@ declare module 'vscode' {
 		 * The number of available tokens that can be used when sending requests
 		 * to the language model.
 		 *
+		 * _Note_ that input- and output-tokens count towards this limit.
+		 *
 		 * @see {@link lm.sendChatRequest}
 		 */
-		// TODO@API is this optional?
-		// TODO@API rename: contextLength?
-		readonly tokens: number;
+		readonly contextLength: number;
 	}
 
 	/**