@gitlab/gitlab-ai-provider 3.3.1 → 3.4.1

package/dist/index.mjs

@@ -38,6 +38,17 @@ var GitLabError = class _GitLabError extends Error {
   isServerError() {
     return this.statusCode !== void 0 && this.statusCode >= 500;
   }
+  /**
+   * Check if this error is a context overflow error (prompt too long).
+   * These errors occur when the conversation exceeds the model's token limit.
+   */
+  isContextOverflowError() {
+    if (this.statusCode !== 400) {
+      return false;
+    }
+    const message = this.message?.toLowerCase() || "";
+    return message.includes("context overflow") || message.includes("prompt is too long") || message.includes("prompt too long") || message.includes("tokens") && message.includes("maximum");
+  }
 };
 
 // src/gitlab-direct-access.ts
@@ -181,11 +192,15 @@ var GitLabAnthropicLanguageModel = class {
   async getAnthropicClient(forceRefresh = false) {
     const tokenData = await this.directAccessClient.getDirectAccessToken(forceRefresh);
     const { "x-api-key": _removed, ...filteredHeaders } = tokenData.headers;
+    const mergedHeaders = {
+      ...filteredHeaders,
+      ...this.config.aiGatewayHeaders
+    };
     this.anthropicClient = new Anthropic({
       apiKey: null,
       authToken: tokenData.token,
       baseURL: this.directAccessClient.getAnthropicProxyUrl(),
-      defaultHeaders: filteredHeaders
+      defaultHeaders: mergedHeaders
     });
     return this.anthropicClient;
   }
@@ -204,6 +219,21 @@ var GitLabAnthropicLanguageModel = class {
     }
     return false;
   }
+  /**
+   * Check if an error is a context overflow error (prompt too long)
+   * These should NOT trigger token refresh and should be reported to the user.
+   */
+  isContextOverflowError(error) {
+    if (error instanceof Anthropic.APIError) {
+      if (error.status === 400) {
+        const message = error.message?.toLowerCase() || "";
+        if (message.includes("prompt is too long") || message.includes("prompt too long") || message.includes("tokens") && message.includes("maximum")) {
+          return true;
+        }
+      }
+    }
+    return false;
+  }
   /**
    * Convert AI SDK tools to Anthropic tool format
    */
@@ -381,6 +411,14 @@ var GitLabAnthropicLanguageModel = class {
         warnings: []
       };
     } catch (error) {
+      if (this.isContextOverflowError(error)) {
+        const apiError = error;
+        throw new GitLabError({
+          message: `Context overflow: ${apiError.message}. Please start a new session or use /compact to reduce context.`,
+          statusCode: 400,
+          cause: error
+        });
+      }
       if (!isRetry && this.isTokenError(error)) {
         this.directAccessClient.invalidateToken();
         return this.doGenerateWithRetry(options, true);
@@ -388,6 +426,7 @@ var GitLabAnthropicLanguageModel = class {
       if (error instanceof Anthropic.APIError) {
         throw new GitLabError({
           message: `Anthropic API error: ${error.message}`,
+          statusCode: error.status,
           cause: error
         });
       }
@@ -568,6 +607,19 @@ var GitLabAnthropicLanguageModel = class {
               });
             }
           }
+          if (self.isContextOverflowError(error)) {
+            const apiError = error;
+            controller.enqueue({
+              type: "error",
+              error: new GitLabError({
+                message: `Context overflow: ${apiError.message}. Please start a new session or use /compact to reduce context.`,
+                statusCode: 400,
+                cause: error
+              })
+            });
+            controller.close();
+            return;
+          }
           if (!isRetry && self.isTokenError(error)) {
             self.directAccessClient.invalidateToken();
             controller.enqueue({
@@ -585,6 +637,7 @@ var GitLabAnthropicLanguageModel = class {
               type: "error",
               error: new GitLabError({
                 message: `Anthropic API error: ${error.message}`,
+                statusCode: error.status,
                 cause: error
               })
             });
@@ -686,10 +739,14 @@ var GitLabOpenAILanguageModel = class {
   async getOpenAIClient(forceRefresh = false) {
     const tokenData = await this.directAccessClient.getDirectAccessToken(forceRefresh);
     const { "x-api-key": _removed, ...filteredHeaders } = tokenData.headers;
+    const mergedHeaders = {
+      ...filteredHeaders,
+      ...this.config.aiGatewayHeaders
+    };
     this.openaiClient = new OpenAI({
       apiKey: tokenData.token,
       baseURL: this.directAccessClient.getOpenAIProxyUrl(),
-      defaultHeaders: filteredHeaders
+      defaultHeaders: mergedHeaders
     });
     return this.openaiClient;
   }
@@ -705,6 +762,21 @@ var GitLabOpenAILanguageModel = class {
     }
     return false;
   }
+  /**
+   * Check if an error is a context overflow error (prompt too long)
+   * These should NOT trigger token refresh and should be reported to the user.
+   */
+  isContextOverflowError(error) {
+    if (error instanceof OpenAI.APIError) {
+      if (error.status === 400) {
+        const message = error.message?.toLowerCase() || "";
+        if (message.includes("prompt is too long") || message.includes("prompt too long") || message.includes("tokens") && message.includes("maximum")) {
+          return true;
+        }
+      }
+    }
+    return false;
+  }
   convertTools(tools) {
     if (!tools || tools.length === 0) {
       return void 0;
@@ -985,6 +1057,14 @@ var GitLabOpenAILanguageModel = class {
         warnings: []
       };
     } catch (error) {
+      if (this.isContextOverflowError(error)) {
+        const apiError = error;
+        throw new GitLabError({
+          message: `Context overflow: ${apiError.message}. Please start a new session or use /compact to reduce context.`,
+          statusCode: 400,
+          cause: error
+        });
+      }
       if (!isRetry && this.isTokenError(error)) {
         this.directAccessClient.invalidateToken();
         return this.doGenerateWithChatApi(options, true);
@@ -992,6 +1072,7 @@ var GitLabOpenAILanguageModel = class {
       if (error instanceof OpenAI.APIError) {
         throw new GitLabError({
           message: `OpenAI API error: ${error.message}`,
+          statusCode: error.status,
           cause: error
         });
       }
@@ -1047,6 +1128,14 @@ var GitLabOpenAILanguageModel = class {
         warnings: []
       };
     } catch (error) {
+      if (this.isContextOverflowError(error)) {
+        const apiError = error;
+        throw new GitLabError({
+          message: `Context overflow: ${apiError.message}. Please start a new session or use /compact to reduce context.`,
+          statusCode: 400,
+          cause: error
+        });
+      }
       if (!isRetry && this.isTokenError(error)) {
         this.directAccessClient.invalidateToken();
         return this.doGenerateWithResponsesApi(options, true);
@@ -1054,6 +1143,7 @@ var GitLabOpenAILanguageModel = class {
       if (error instanceof OpenAI.APIError) {
         throw new GitLabError({
           message: `OpenAI API error: ${error.message}`,
+          statusCode: error.status,
           cause: error
         });
       }
@@ -1172,6 +1262,19 @@ var GitLabOpenAILanguageModel = class {
           controller.enqueue({ type: "finish", finishReason, usage });
           controller.close();
         } catch (error) {
+          if (self.isContextOverflowError(error)) {
+            const apiError = error;
+            controller.enqueue({
+              type: "error",
+              error: new GitLabError({
+                message: `Context overflow: ${apiError.message}. Please start a new session or use /compact to reduce context.`,
+                statusCode: 400,
+                cause: error
+              })
+            });
+            controller.close();
+            return;
+          }
           if (!isRetry && self.isTokenError(error)) {
             self.directAccessClient.invalidateToken();
             controller.enqueue({
@@ -1186,6 +1289,7 @@ var GitLabOpenAILanguageModel = class {
               type: "error",
               error: new GitLabError({
                 message: `OpenAI API error: ${error.message}`,
+                statusCode: error.status,
                 cause: error
               })
             });
@@ -1312,6 +1416,19 @@ var GitLabOpenAILanguageModel = class {
           controller.enqueue({ type: "finish", finishReason, usage });
           controller.close();
         } catch (error) {
+          if (self.isContextOverflowError(error)) {
+            const apiError = error;
+            controller.enqueue({
+              type: "error",
+              error: new GitLabError({
+                message: `Context overflow: ${apiError.message}. Please start a new session or use /compact to reduce context.`,
+                statusCode: 400,
+                cause: error
+              })
+            });
+            controller.close();
+            return;
+          }
           if (!isRetry && self.isTokenError(error)) {
             self.directAccessClient.invalidateToken();
             controller.enqueue({
@@ -1326,6 +1443,7 @@ var GitLabOpenAILanguageModel = class {
               type: "error",
               error: new GitLabError({
                 message: `OpenAI API error: ${error.message}`,
+                statusCode: error.status,
                 cause: error
               })
             });
@@ -1502,11 +1620,13 @@ var GitLabOAuthManager = class {
   }
 };
 
+// src/version.ts
+var VERSION = true ? "3.4.0" : "0.0.0-dev";
+
 // src/gitlab-provider.ts
 import * as fs from "fs";
 import * as path from "path";
 import * as os from "os";
-var VERSION = "0.0.1";
 function getOpenCodeAuthPath() {
   const homeDir = os.homedir();
   const xdgDataHome = process.env.XDG_DATA_HOME;
@@ -1656,6 +1776,14 @@ function createGitLab(options = {}) {
       ...options.featureFlags,
       ...agenticOptions?.featureFlags
     };
+    const defaultAiGatewayHeaders = {
+      "User-Agent": `gitlab-ai-provider/${VERSION}`
+    };
+    const aiGatewayHeaders = {
+      ...defaultAiGatewayHeaders,
+      ...options.aiGatewayHeaders,
+      ...agenticOptions?.aiGatewayHeaders
+    };
     const baseConfig = {
       provider: `${providerName}.agentic`,
       instanceUrl,
@@ -1664,7 +1792,8 @@ function createGitLab(options = {}) {
       fetch: options.fetch,
       maxTokens: agenticOptions?.maxTokens,
       featureFlags,
-      aiGatewayUrl: options.aiGatewayUrl
+      aiGatewayUrl: options.aiGatewayUrl,
+      aiGatewayHeaders
     };
     if (mapping.provider === "openai") {
       return new GitLabOpenAILanguageModel(modelId, {
@@ -1984,6 +2113,7 @@ export {
   MODEL_MAPPINGS,
   OAUTH_SCOPES,
   TOKEN_EXPIRY_SKEW_MS,
+  VERSION,
   createGitLab,
   getAnthropicModelForModelId,
   getModelMapping,