store model to vscode ml mappings (#596)

* store vscode language mapping in project configuration

* clean imports

* added docs

* add docs

* fix

* lazy update

* review

* more docs

docs/src/assets/vscode-language-models-select.png.txt

@@ -0,0 +1 @@
+A dropdown menu titled 'Pick a Language Chat Model for openai:gpt-4' with several options including 'GPT 3.5 Turbo', 'GPT 4', 'GPT 4 Turbo (2024-01-25 Preview)', and 'GPT 4o (2024-05-13)', with 'GPT 3.5 Turbo' currently highlighted.

docs/src/assets/vscode-language-models.png.txt

@@ -0,0 +1 @@
+Screenshot of a Visual Studio Code interface showing a dropdown menu with options to configure a language model for OpenAI's GPT-4, including an option for Visual Studio Language Chat Models and using a registered LLM such as GitHub Copilot.

docs/src/content/docs/getting-started/configuration.mdx

@@ -11,11 +11,11 @@ import { Steps } from "@astrojs/starlight/components"
 import { Tabs, TabItem } from "@astrojs/starlight/components"
 import { Image } from "astro:assets"
 
-import insidersSrc from "../../../assets/vscode-insiders.png"
-import insidersAlt from "../../../assets/vscode-insiders.png.txt?raw"
+import lmSrc from "../../../assets/vscode-language-models.png"
+import lmAlt from "../../../assets/vscode-language-models.png.txt?raw"
 
-import selectLLMSrc from "../../../assets/vscode-select-llm.png"
-import selectLLMAlt from "../../../assets/vscode-select-llm.png.txt?raw"
+import lmSelectSrc from "../../../assets/vscode-language-models-select.png"
+import lmSelectAlt from "../../../assets/vscode-language-models-select.png.txt?raw"
 
 You will need to configure the LLM connection and authorizion secrets.
 
@@ -269,6 +269,42 @@ script({
 
 </Steps>
 
+## GitHub Copilot Models
+
+If you have access to **GitHub Copilot in Visual Studio Code**, 
+GenAIScript will be able to leverage those [language models](https://code.visualstudio.com/api/extension-guides/language-model) as well.
+
+This mode is useful to run your scripts without having a separate LLM provider or local LLMs. However, those models are  not available from the command line
+and have additional limitations and rate limiting defined by the GitHub Copilot platform.
+
+There is no configuration needed as long as you have GitHub Copilot installed and configured in Visual Studio Code.
+
+<Steps>
+
+<ol>
+
+<li>run your script</li>
+<li> 
+select the **Visual Studio Code Language Chat Models** option when configuring the model
+
+<Image src={lmSrc} alt={lmAlt} loading="lazy" />
+
+(This step is skipped if you already have mappings in your settings)
+
+</li>
+<li>
+select the best chat model that matches the one you have in your script
+
+<Image src={lmSelectSrc} alt={lmSelectAlt} loading="lazy" />
+
+</li>
+
+</ol>
+
+</Steps>
+
+The mapping of GenAIScript model names to Visual Studio Models is stored in the settings.
+
 ## Local Models
 
 There are many projects that allow you to run models locally on your machine,

docs/src/content/docs/reference/token.md

@@ -10,7 +10,7 @@ GenAIScript will try to find the connection token from various sources:
 
 -   a `.env` file in the root of your project (VSCode and CLI)
 -   environment variables, typically within your CI/CD environment (CLI only)
--   Visual Studio Language Models (VSCode only)
+-   Visual Studio Language Chat Models (VSCode only)
 
 ## .env file or process environment
 

package.json

@@ -59,7 +59,7 @@
         "gen:licenses": "npx --yes generate-license-file --input ./package.json --output ./THIRD_PARTY_LICENSES.md --overwrite",
         "genai:technical": "cd docs && yarn genai:technical",
         "genai:frontmatter": "cd docs && yarn genai:frontmatter",
-        "genai:alt": "cd docs && yarn genai:alt",
+        "genai:alt": "cd docs && yarn genai:alt-text",
         "genai:test": "node packages/cli/built/genaiscript.cjs run test-gen"
     },
     "release-it": {

packages/cli/src/nodehost.ts

@@ -157,7 +157,6 @@ export class NodeHost implements RuntimeHost {
             tok.token = "Bearer " + this._azureToken
         }
         if (!tok && this.clientLanguageModel) {
-            logVerbose(`model: using client language model`)
             return <LanguageModelConfiguration>{
                 model: modelId,
                 provider: this.clientLanguageModel.id,

packages/cli/src/server.ts

@@ -107,8 +107,9 @@ export async function startServer(options: { port: string }) {
                 // add handler
                 const chatId = randomHex(6)
                 chats[chatId] = async (chunk) => {
-                    if (!responseSoFar) {
-                        trace.itemValue("model", chunk.model)
+                    if (!responseSoFar && chunk.model) {
+                        logVerbose(`visual studio: chat model ${chunk.model}`)
+                        trace.itemValue("chat model", chunk.model)
                         trace.appendContent("\n\n")
                     }
                     trace.appendToken(chunk.chunk)

packages/sample/.vscode/settings.json

@@ -4,5 +4,8 @@
         "openai",
         "outputfilename"
     ],
-    "genaiscript.cli.path": "../cli/built/genaiscript.cjs"
+    "genaiscript.cli.path": "../cli/built/genaiscript.cjs",
+    "genaiscript.languageChatModels": {
+        "openai:gpt-4": "github.copilot-chat/4/gpt-4o-2024-05-13"
+    }
 }

packages/vscode/package.json

@@ -234,6 +234,10 @@
             {
                 "title": "GenAIScript",
                 "properties": {
+                    "genaiscript.languageChatModels": {
+                        "type": "object",
+                        "description": "Mapping from GenAIScript model (openai:gpt-4) to Visual Studio Code Language Chat Model (github...)"
+                    },
                     "genaiscript.diagnostics": {
                         "type": "boolean",
                         "default": false,

packages/vscode/src/lmaccess.ts

@@ -1,8 +1,6 @@
 /* eslint-disable @typescript-eslint/naming-convention */
 import * as vscode from "vscode"
-import { AIRequestOptions, ExtensionState } from "./state"
-import { isApiProposalEnabled } from "./proposals"
-import { LanguageModel } from "../../core/src/chat"
+import { ExtensionState } from "./state"
 import {
     MODEL_PROVIDER_OLLAMA,
     MODEL_PROVIDER_LLAMAFILE,
@@ -34,7 +32,8 @@ async function generateLanguageModelConfiguration(
         return { provider }
     }
 
-    if (Object.keys(state.languageChatModels).length)
+    const languageChatModels = await state.languageChatModels()
+    if (Object.keys(languageChatModels).length)
         return { provider: MODEL_PROVIDER_CLIENT, model: "*" }
 
     const items: (vscode.QuickPickItem & {
@@ -46,8 +45,8 @@ async function generateLanguageModelConfiguration(
         const models = await vscode.lm.selectChatModels()
         if (models.length)
             items.push({
-                label: "Visual Studio Language Models",
-                detail: `Use a registered Language Model (e.g. GitHub Copilot).`,
+                label: "Visual Studio Language Chat Models",
+                detail: `Use a registered LLM such as GitHub Copilot.`,
                 model: "*",
                 provider: MODEL_PROVIDER_CLIENT,
             })
@@ -104,8 +103,8 @@ async function pickChatModel(
     model: string
 ): Promise<vscode.LanguageModelChat> {
     const chatModels = await vscode.lm.selectChatModels()
-
-    const chatModelId = state.languageChatModels[model]
+    const languageChatModels = await state.languageChatModels()
+    const chatModelId = languageChatModels[model]
     let chatModel = chatModelId && chatModels.find((m) => m.id === chatModelId)
     if (!chatModel) {
         const items: (vscode.QuickPickItem & {
@@ -117,10 +116,10 @@ async function pickChatModel(
             chatModel,
         }))
         const res = await vscode.window.showQuickPick(items, {
-            title: `Pick a Chat Model for ${model}`,
+            title: `Pick a Language Chat Model for ${model}`,
         })
         chatModel = res?.chatModel
-        if (chatModel) state.languageChatModels[model] = chatModel.id
+        if (chatModel) await state.updateLanguageChatModels(model, chatModel.id)
     }
     return chatModel
 }

packages/vscode/src/servermanager.ts

@@ -39,7 +39,7 @@ export class TerminalServerManager implements ServerManager {
         )
         subscriptions.push(
             vscode.workspace.onDidChangeConfiguration((e) => {
-                if (e.affectsConfiguration(TOOL_ID)) this.close()
+                if (e.affectsConfiguration(TOOL_ID + ".cli")) this.close()
             })
         )
 

packages/vscode/src/state.ts

@@ -107,8 +107,6 @@ export class ExtensionState extends EventTarget {
         AIRequestSnapshot
     > = undefined
     readonly output: vscode.LogOutputChannel
-    // modelid -> vscode language mode id
-    languageChatModels: Record<string, string> = {}
 
     constructor(public readonly context: ExtensionContext) {
         super()
@@ -138,15 +136,26 @@ export class ExtensionState extends EventTarget {
                 subscriptions
             )
         )
-        if (
-            typeof vscode.lm !== "undefined" &&
-            typeof vscode.lm.onDidChangeChatModels === "function"
-        )
-            subscriptions.push(
-                vscode.lm.onDidChangeChatModels(
-                    () => (this.languageChatModels = {})
-                )
-            )
+    }
+
+    async updateLanguageChatModels(model: string, chatModel: string) {
+        const res = await this.languageChatModels()
+        if (res[model] !== chatModel) {
+            if (chatModel === undefined) delete res[model]
+            else res[model] = chatModel
+            const config = vscode.workspace.getConfiguration(TOOL_ID)
+            await config.update("languageChatModels", res)
+        }
+    }
+
+    async languageChatModels() {
+        const config = vscode.workspace.getConfiguration(TOOL_ID)
+        const res =
+            ((await config.get("languageChatModels")) as Record<
+                string,
+                string
+            >) || {}
+        return res
     }
 
     private async saveScripts() {