Add ability to use a dev endpoint for auto-model (#3038)

extensions/ql-vscode/src/config.ts

@@ -706,6 +706,10 @@ const LLM_GENERATION_BATCH_SIZE = new Setting(
   "llmGenerationBatchSize",
   MODEL_SETTING,
 );
+const LLM_GENERATION_DEV_ENDPOINT = new Setting(
+  "llmGenerationDevEndpoint",
+  MODEL_SETTING,
+);
 const EXTENSIONS_DIRECTORY = new Setting("extensionsDirectory", MODEL_SETTING);
 const ENABLE_RUBY = new Setting("enableRuby", MODEL_SETTING);
 
@@ -738,6 +742,14 @@ export class ModelConfigListener extends ConfigListener implements ModelConfig {
     return LLM_GENERATION_BATCH_SIZE.getValue<number | null>() || 5;
   }
 
+  /**
+   * The URL of the endpoint to use for LLM generation. This should only be set
+   * if you want to test against a dev server.
+   */
+  public get llmGenerationDevEndpoint(): string | undefined {
+    return LLM_GENERATION_DEV_ENDPOINT.getValue<string | undefined>();
+  }
+
   public getExtensionsDirectory(languageId: string): string | undefined {
     return EXTENSIONS_DIRECTORY.getValue<string>({
       languageId,

extensions/ql-vscode/src/model-editor/auto-model-api.ts

@@ -1,5 +1,7 @@
 import { Credentials } from "../common/authentication";
 import { OctokitResponse } from "@octokit/types";
+import fetch from "node-fetch";
+import { ModelConfigListener } from "../config";
 
 export enum AutomodelMode {
   Unspecified = "AUTOMODEL_MODE_UNSPECIFIED",
@@ -20,15 +22,44 @@ export interface ModelResponse {
 export async function autoModel(
   credentials: Credentials,
   request: ModelRequest,
+  modelingConfig: ModelConfigListener,
 ): Promise<ModelResponse> {
-  const octokit = await credentials.getOctokit();
+  const devEndpoint = modelingConfig.llmGenerationDevEndpoint;
+  if (devEndpoint) {
+    return callAutoModelDevEndpoint(devEndpoint, request);
+  } else {
+    const octokit = await credentials.getOctokit();
 
-  const response: OctokitResponse<ModelResponse> = await octokit.request(
-    "POST /repos/github/codeql/code-scanning/codeql/auto-model",
-    {
-      data: request,
-    },
-  );
+    const response: OctokitResponse<ModelResponse> = await octokit.request(
+      "POST /repos/github/codeql/code-scanning/codeql/auto-model",
+      {
+        data: request,
+      },
+    );
 
-  return response.data;
+    return response.data;
+  }
+}
+
+async function callAutoModelDevEndpoint(
+  endpoint: string,
+  request: ModelRequest,
+): Promise<ModelResponse> {
+  const json = JSON.stringify(request);
+  const response = await fetch(endpoint, {
+    method: "POST",
+    headers: {
+      "Content-Type": "application/json",
+    },
+    body: json,
+  });
+
+  if (!response.ok) {
+    throw new Error(
+      `Error calling auto-model API: ${response.status} ${response.statusText}`,
+    );
+  }
+
+  const data = await response.json();
+  return data as ModelResponse;
 }

extensions/ql-vscode/src/model-editor/auto-modeler.ts

@@ -238,7 +238,7 @@ export class AutoModeler {
     request: ModelRequest,
   ): Promise<ModelResponse | null> {
     try {
-      return await autoModel(this.app.credentials, request);
+      return await autoModel(this.app.credentials, request, this.modelConfig);
     } catch (e) {
       if (e instanceof RequestError && e.status === 429) {
         void showAndLogExceptionWithTelemetry(