Class AnthropicEngine

src/engine/anthropic.ts:16–75

interface AnthropicConfig extends AiEngineConfig {}

export class AnthropicEngine implements AiEngine {
  config: AnthropicConfig;
  client: AnthropicClient;

  constructor(config) {
    this.config = config;
    const clientOptions: any = { apiKey: this.config.apiKey };

    const proxy = config.proxy;
    if (proxy) {
      clientOptions.httpAgent = new HttpsProxyAgent(proxy);
    }

    this.client = new AnthropicClient(clientOptions);
  }

  public generateCommitMessage = async (
    messages: Array<OpenAI.Chat.Completions.ChatCompletionMessageParam>
  ): Promise<string | undefined> => {
    const systemMessage = messages.find((msg) => msg.role === 'system')
      ?.content as string;
    const restMessages = messages.filter(
      (msg) => msg.role !== 'system'
    ) as MessageParam[];

    const params: MessageCreateParamsNonStreaming = {
      model: this.config.model,
      system: systemMessage,
      messages: restMessages,
      temperature: 0,
      max_tokens: this.config.maxTokensOutput
    };

    // Anthropic rejects simultaneous temperature and top_p on Claude 4.5+
    if (!/claude.*-4-(?:5|[6-9])/.test(params.model)) {
      params.top_p = 0.1;
    }

    try {
      const REQUEST_TOKENS = messages
        .map((msg) => tokenCount(msg.content as string) + 4)
        .reduce((a, b) => a + b, 0);

      if (
        REQUEST_TOKENS >
        this.config.maxTokensInput - this.config.maxTokensOutput
      ) {
        throw new Error(GenerateCommitMessageErrorEnum.tooMuchTokens);
      }

      const data = await this.client.messages.create(params);

      const message = data?.content[0].text;
      let content = message;
      return removeContentTags(content, 'think');
    } catch (error) {
      throw normalizeEngineError(error, 'anthropic', this.config.model);
    }