Skip to content

Voice-to-text dictation app with local (Nvidia Parakeet/Whisper) and cloud models (BYOK). Privacy-first and available cross-platform.

Merge remote-tracking branch 'origin/main' into chore/update-reasoning-model-catalog

Gabriel Stein5b186e781cea
diff --cc src/services/ReasoningService.ts
index eb742d7e4,e1c822d26..01702128b
--- a/src/services/ReasoningService.ts
+++ b/src/services/ReasoningService.ts
@@@ -176,9 -168,134 +168,134 @@@ class ReasoningService extends BaseReas
return apiKey;
}
+ /**
+ * Shared helper for OpenAI-compatible Chat Completions API calls.
+ * Used by Groq and as fallback for OpenAI.
+ */
+ private async callChatCompletionsApi(
+ endpoint: string,
+ apiKey: string,
+ model: string,
+ text: string,
+ agentName: string | null,
+ config: ReasoningConfig,
+ providerName: string
+ ): Promise<string> {
+ const systemPrompt = "You are a dictation assistant. Clean up text by fixing grammar and punctuation. Output ONLY the cleaned text without any explanations, options, or commentary.";
+ const userPrompt = this.getReasoningPrompt(text, agentName, config);
+
+ const messages = [
+ { role: "system", content: systemPrompt },
+ { role: "user", content: userPrompt }
+ ];
+
+ const requestBody: any = {
+ model,
+ messages,
+ temperature: config.temperature ?? 0.3,
+ max_tokens: config.maxTokens || Math.max(
+ 4096,
+ this.calculateMaxTokens(
+ text.length,
+ TOKEN_LIMITS.MIN_TOKENS,
+ TOKEN_LIMITS.MAX_TOKENS,
+ TOKEN_LIMITS.TOKEN_MULTIPLIER
+ )
+ ),
+ };
+
+ logger.logReasoning(`${providerName.toUpperCase()}_REQUEST`, {
+ endpoint,
+ model,
+ hasApiKey: !!apiKey,
+ requestBody: JSON.stringify(requestBody).substring(0, 200)
+ });
+
+ const response = await withRetry(
+ async () => {
+ const res = await fetch(endpoint, {
+ method: "POST",
+ headers: {
+ "Content-Type": "application/json",
+ Authorization: `Bearer ${apiKey}`,
+ },
+ body: JSON.stringify(requestBody),
+ });
+
+ if (!res.ok) {
+ const errorText = await res.text();
+ let errorData: any = { error: res.statusText };
+
+ try {
+ errorData = JSON.parse(errorText);
+ } catch {
+ errorData = { error: errorText || res.statusText };
+ }
+
+ logger.logReasoning(`${providerName.toUpperCase()}_API_ERROR_DETAIL`, {
+ status: res.status,
+ statusText: res.statusText,
+ error: errorData,
+ errorMessage: errorData.error?.message || errorData.message || errorData.error,
+ fullResponse: errorText.substring(0, 500)
+ });
+
+ const errorMessage = errorData.error?.message || errorData.message || errorData.error || `${providerName} API error: ${res.status}`;
+ throw new Error(errorMessage);
+ }
+
+ const jsonResponse = await res.json();
+
+ logger.logReasoning(`${providerName.toUpperCase()}_RAW_RESPONSE`, {
+ hasResponse: !!jsonResponse,
+ responseKeys: jsonResponse ? Object.keys(jsonResponse) : [],
+ hasChoices: !!jsonResponse?.choices,
+ choicesLength: jsonResponse?.choices?.length || 0,
+ fullResponse: JSON.stringify(jsonResponse).substring(0, 500)
+ });
+
+ return jsonResponse;
+ },
+ createApiRetryStrategy()
+ );
+
+ // Extract text from Chat Completions format
+ if (!response.choices || !response.choices[0]) {
+ logger.logReasoning(`${providerName.toUpperCase()}_RESPONSE_ERROR`, {
+ model,
+ response: JSON.stringify(response).substring(0, 500),
+ hasChoices: !!response.choices,
+ choicesCount: response.choices?.length || 0
+ });
+ throw new Error(`Invalid response structure from ${providerName} API`);
+ }
+
+ const choice = response.choices[0];
+ const responseText = choice.message?.content?.trim() || "";
+
+ if (!responseText) {
+ logger.logReasoning(`${providerName.toUpperCase()}_EMPTY_RESPONSE`, {
+ model,
+ finishReason: choice.finish_reason,
+ hasMessage: !!choice.message,
+ response: JSON.stringify(choice).substring(0, 500)
+ });
+ throw new Error(`${providerName} returned empty response`);
+ }
+
+ logger.logReasoning(`${providerName.toUpperCase()}_RESPONSE`, {
+ model,
+ responseLength: responseText.length,
+ tokensUsed: response.usage?.total_tokens || 0,
+ success: true
+ });
+
+ return responseText;
+ }
+
async processText(
text: string,
- model: string = "gpt-4o-mini",
+ model: string = "",
agentName: string | null = null,
config: ReasoningConfig = {}
): Promise<string> {
@@@ -208,17 -321,20 +325,20 @@@
switch (provider) {
case "openai":
- result = await this.processWithOpenAI(text, model, agentName, config);
+ result = await this.processWithOpenAI(text, trimmedModel, agentName, config);
break;
case "anthropic":
- result = await this.processWithAnthropic(text, model, agentName, config);
+ result = await this.processWithAnthropic(text, trimmedModel, agentName, config);
break;
case "local":
- result = await this.processWithLocal(text, model, agentName, config);
+ result = await this.processWithLocal(text, trimmedModel, agentName, config);
break;
case "gemini":
- result = await this.processWithGemini(text, model, agentName, config);
+ result = await this.processWithGemini(text, trimmedModel, agentName, config);
break;
+ case "groq":
+ result = await this.processWithGroq(text, model, agentName, config);
+ break;
default:
throw new Error(`Unsupported reasoning provider: ${provider}`);
}