continuedev/continue · error · Error

Failed to parse config.json: ${e}

Error message

Failed to parse config.json: ${e}

What it means

Vertex AI express mode (API key auth) only supports Gemini endpoints; the adapter detects this combination during non-streaming chat and rejects it before making a request. Anthropic-on-Vertex and Mistral-on-Vertex require service-account OAuth.

Source

Thrown at core/config/load.ts:124

    return a.name === b.name && a.params?.url === b.params?.url;
  },
  slashCommands: (a: any, b: any) => a.name === b.name,
  customCommands: (a: any, b: any) => a.name === b.name,
};

function loadSerializedConfig(
  workspaceConfigs: ContinueRcJson[],
  ideSettings: IdeSettings,
  ideType: IdeType,
  overrideConfigJson: SerializedContinueConfig | undefined,
  ide: IDE,
): ConfigResult<SerializedContinueConfig> {
  let config: SerializedContinueConfig = overrideConfigJson!;
  if (!config) {
    try {
      config = resolveSerializedConfig(getConfigJsonPath());
    } catch (e) {
      throw new Error(`Failed to parse config.json: ${e}`);
    }
  }

  const errors = validateConfig(config);

  if (errors?.some((error) => error.fatal)) {
    return {
      errors,
      config: undefined,
      configLoadInterrupted: true,
    };
  }

  if (config.allowAnonymousTelemetry === undefined) {
    config.allowAnonymousTelemetry = true;
  }

  if (ideSettings.remoteConfigServerUrl) {

View on GitHub (pinned to 5522c6f44c)

Solutions

1. Switch to service-account auth: remove apiKey and set keyJson/keyFile (or ADC) for non-gemini models
2. Or keep a gemini-* model when using express apiKey mode
3. Configure separate adapter instances per auth mode if you need both

Example fix

// before
new VertexAIApi({ apiKey: process.env.VERTEX_API_KEY, ... });
await api.chatCompletionNonStream({ model: 'claude-3-5-sonnet@...', ... }, signal);

// after
new VertexAIApi({ keyJson: SA_JSON, ... });
await api.chatCompletionNonStream({ model: 'claude-3-5-sonnet@...', ... }, signal);

const expressOk = (model: string, hasApiKey: boolean) => !hasApiKey || /gemini/i.test(model);

const isGeminiModel = (m: string): boolean => /gemini/i.test(m);

try { ... } catch (e) { if ((e as Error).message.includes('express (apiKey) mode')) { /* switch to service-account auth or a gemini model */ } throw e; }

When it happens

Trigger: config.apiKey is set AND body.model resolves to a non-gemini provider (claude-* or mistral model) in chatCompletionNonStream.

Common situations: Developer reuses an OpenAI-style API key config then switches the model to claude-3-5-sonnet or codestral; or assumes the Vertex API key authenticates all publisher models.

Understand the failure class

• Parsing and encoding errors: unexpected token, malformed input — why parsers reject input and how to find the real culprit.

Related errors

• MCP Connection ${serverId} not found
• URL is not defined in params
• Response body is null
• Profile ${profileId} not found
• No workspace directories found. Make sure you've opened a fo

AI-assisted analysis of continuedev/continue@5522c6f44c (2026-08-27). Data as JSON: /api/errors/fd8531b9ad777663. Report an issue: GitHub.