Skip to content

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

refactor(reasoning): unify dialect fallback and error extraction

Gabriel Stein8c4aa973553d
diff --git a/src/services/ReasoningService.ts b/src/services/ReasoningService.ts
index 32c714e77..8d9165c02 100644
--- a/src/services/ReasoningService.ts
+++ b/src/services/ReasoningService.ts
@@ -443,14 +443,7 @@ class ReasoningService extends BaseReasoningService {
}
// A known endpoint host knows its own request shape better than the model id does.
- const endpointDialect = detectEndpointDialect(endpoint);
- const registryConfig = getOpenAiApiConfig(model, provider);
- const apiConfig = endpointDialect
- ? {
- tokenParam: endpointDialect.tokenParam,
- supportsTemperature: endpointDialect.supportsTemperature,
- }
- : registryConfig;
+ const apiConfig = detectEndpointDialect(endpoint) ?? getOpenAiApiConfig(model, provider);
const useOldTokenParam = isLocalProvider || isLanChat || provider === "groq";
const requestBody: Record<string, unknown> = {
@@ -659,7 +652,7 @@ class ReasoningService extends BaseReasoningService {
disableThinking: openrouterDisableThinking,
});
- const apiConfig = getOpenAiApiConfig(model, provider);
+ const apiConfig = detectEndpointDialect(baseURL) ?? getOpenAiApiConfig(model, provider);
const modelDef = getCloudModel(model);
const userSuppressesThinking = config.disableThinking === true && !!modelDef?.supportsThinking;
const needsGroqDisableThinking =
diff --git a/src/services/ai/inferenceProviders/gemini.ts b/src/services/ai/inferenceProviders/gemini.ts
index ee5152186..17a3d1968 100644
--- a/src/services/ai/inferenceProviders/gemini.ts
+++ b/src/services/ai/inferenceProviders/gemini.ts
@@ -3,6 +3,7 @@ import { getCloudModel } from "../../../models/ModelRegistry";
import { withRetry, createApiRetryStrategy, httpError } from "../../../utils/retry";
import { API_ENDPOINTS, TOKEN_LIMITS } from "../../../config/constants";
import { wrapCleanupTranscript } from "../../../config/prompts";
+import { extractApiErrorMessage } from "../apiErrorMessage";
import logger from "../../../utils/logger";
interface GeminiResponse {
@@ -92,11 +93,7 @@ export const geminiProvider: InferenceProvider = {
fullResponse: errorText.substring(0, 500),
});
- const errMsg =
- (typeof errorData.error === "object" && errorData.error?.message) ||
- errorData.message ||
- (typeof errorData.error === "string" ? errorData.error : null) ||
- `Gemini API error: ${res.status}`;
+ const errMsg = extractApiErrorMessage(errorData, `Gemini API error: ${res.status}`);
throw httpError(errMsg, res.status);
}
diff --git a/src/services/ai/thinkingSuppressionDialects.ts b/src/services/ai/thinkingSuppressionDialects.ts
index 7e6ae588f..ff754e33f 100644
--- a/src/services/ai/thinkingSuppressionDialects.ts
+++ b/src/services/ai/thinkingSuppressionDialects.ts
@@ -4,8 +4,8 @@
*/
export interface EndpointDialect {
key: "mistral";
- tokenParam: "max_tokens";
- supportsTemperature: true;
+ tokenParam: "max_tokens" | "max_completion_tokens";
+ supportsTemperature: boolean;
}
/** Custom endpoints that need their own request shape, recognised by host. */