Voice-to-text dictation app with local (Nvidia Parakeet/Whisper) and cloud models (BYOK). Privacy-first and available cross-platform.
JavaScript aianthropiccross-platformgeminigroqlinuxmacosnvidiaopen-sourceopenaiparakeetspeech-to-texttranscribewhisperwindows
// Split query/hash so path-suffix stripping and later joins operate on the
// path only. Provider docs and Azure/gateway pastes often include ?api-version=
// (or similar) on a full /chat/completions URL (#1309).
function splitUrlDecorators(value: string): { path: string; query: string; hash: string } {
let path = value;
let hash = "";
const hashIndex = path.indexOf("#");
if (hashIndex >= 0) {
hash = path.slice(hashIndex);
path = path.slice(0, hashIndex);
}
let query = "";
const queryIndex = path.indexOf("?");
if (queryIndex >= 0) {
query = path.slice(queryIndex);
path = path.slice(0, queryIndex);
}
return { path, query, hash };
}
function joinUrlDecorators(path: string, query: string, hash: string): string {
return `${path}${query}${hash}`;
}
// API Configuration helpers
export const normalizeBaseUrl = (value?: string | null): string => {
if (!value) return "";
const trimmed = value.trim();
if (!trimmed) return "";
const { path: rawPath, query, hash } = splitUrlDecorators(trimmed);
let normalized = rawPath;
// Remove common API endpoint suffixes to get the base URL
const suffixReplacements: Array<[RegExp, string]> = [
[/\/v1\/chat\/completions$/i, "/v1"],
[/\/chat\/completions$/i, ""],
[/\/v1\/responses$/i, "/v1"],
[/\/responses$/i, ""],
[/\/v1\/models$/i, "/v1"],
[/\/models$/i, ""],
[/\/v1\/audio\/transcriptions$/i, "/v1"],
[/\/audio\/transcriptions$/i, ""],
[/\/v1\/audio\/translations$/i, "/v1"],
[/\/audio\/translations$/i, ""],
];
for (const [pattern, replacement] of suffixReplacements) {
if (pattern.test(normalized)) {
normalized = normalized.replace(pattern, replacement).replace(/\/+$/, "");
}
}
return joinUrlDecorators(normalized.replace(/\/+$/, ""), query, hash);
};
export const buildApiUrl = (base: string, path: string): string => {
const normalizedBase = normalizeBaseUrl(base) || "https://api.openai.com/v1";
if (!path) {
return normalizedBase;
}
const normalizedPath = path.startsWith("/") ? path : `/${path}`;
const { path: originAndPath, query, hash } = splitUrlDecorators(normalizedBase);
return joinUrlDecorators(`${originAndPath}${normalizedPath}`, query, hash);
};
export const ensureV1Suffix = (base: string): string => {
if (!base) return base;
const normalized = normalizeBaseUrl(base) || base;
const { path, query, hash } = splitUrlDecorators(normalized);
return joinUrlDecorators(path.endsWith("/v1") ? path : `${path}/v1`, query, hash);
};
// Ordered bases to try when listing models from an OpenAI-compatible server.
// Self-hosted servers (LM Studio, Ollama, vLLM) serve the API under /v1 even
// when users enter the bare origin, and LM Studio's native REST base
// (/api/v1 or /api/v0) has its OpenAI-compatible sibling at /v1.
export const getModelListBaseCandidates = (base: string): string[] => {
const normalized = normalizeBaseUrl(base);
if (!normalized) return [];
const { path, query, hash } = splitUrlDecorators(normalized);
const nativeApiMatch = path.match(/^(.+?)\/api\/v[01]$/i);
if (nativeApiMatch) {
return [normalized, joinUrlDecorators(`${nativeApiMatch[1]}/v1`, query, hash)];
}
const withV1 = ensureV1Suffix(normalized);
return withV1 === normalized ? [normalized] : [normalized, withV1];
};
const env = (typeof import.meta !== "undefined" && (import.meta as any).env) || {};
const computeBaseUrl = (candidates: Array<string | undefined>, fallback: string): string => {
for (const candidate of candidates) {
const normalized = normalizeBaseUrl(candidate);
if (normalized) {
return normalized;
}
}
return fallback;
};
const DEFAULT_OPENAI_BASE = computeBaseUrl(
[env.OPENWHISPR_OPENAI_BASE_URL as string | undefined, env.OPENAI_BASE_URL as string | undefined],
"https://api.openai.com/v1"
);
const DEFAULT_TRANSCRIPTION_BASE = computeBaseUrl(
[
env.OPENWHISPR_TRANSCRIPTION_BASE_URL as string | undefined,
env.WHISPER_BASE_URL as string | undefined,
],
DEFAULT_OPENAI_BASE
);
export const API_ENDPOINTS = {
OPENAI_BASE: DEFAULT_OPENAI_BASE,
OPENAI: buildApiUrl(DEFAULT_OPENAI_BASE, "/responses"),
OPENAI_MODELS: buildApiUrl(DEFAULT_OPENAI_BASE, "/models"),
ANTHROPIC: "https://api.anthropic.com/v1/messages",
GEMINI: "https://generativelanguage.googleapis.com/v1beta",
GROQ_BASE: "https://api.groq.com/openai/v1",
CORTI_MODELS_BASE: "https://ai.eu.corti.app/v1",
OPENROUTER_BASE: "https://openrouter.ai/api/v1",
TRANSCRIPTION_BASE: DEFAULT_TRANSCRIPTION_BASE,
TRANSCRIPTION: buildApiUrl(DEFAULT_TRANSCRIPTION_BASE, "/audio/transcriptions"),
} as const;
export const API_VERSIONS = {
ANTHROPIC: "2023-06-01",
GEMINI: "v1beta",
} as const;
// Model Configuration
export const MODEL_CONSTRAINTS = {
MIN_FILE_SIZE: 1_000_000, // 1MB minimum for valid model files
MODEL_TEST_TIMEOUT: 5000, // 5 seconds for model validation
INFERENCE_TIMEOUT: 30000, // 30 seconds default (configurable)
} as const;
// List length above which pickers switch to a searchable variant.
export const LIST_SEARCH_THRESHOLD = 12;
// Token Limits
export const TOKEN_LIMITS = {
MIN_TOKENS: 512,
MAX_TOKENS: 2048,
MIN_TOKENS_ANTHROPIC: 100,
MAX_TOKENS_ANTHROPIC: 4096,
TOKEN_MULTIPLIER: 2, // text.length * multiplier
REASONING_CONTEXT_SIZE: 4096,
} as const;
// Cache Configuration
export const CACHE_CONFIG = {
API_KEY_TTL: 3600000, // 1 hour in milliseconds
MODEL_CACHE_SIZE: 3, // Maximum models to keep in memory
AVAILABILITY_CHECK_TTL: 30000, // 30s for accessibility, FFmpeg, tool availability checks
PASTE_DELAY_MS: 50, // Delay before paste simulation to allow clipboard to settle
} as const;
// OpenWhispr Cloud API
export const OPENWHISPR_API_URL = (env.VITE_OPENWHISPR_API_URL as string) || "";
// Retry Configuration
export const RETRY_CONFIG = {
MAX_RETRIES: 3,
INITIAL_DELAY: 1000, // 1 second
MAX_DELAY: 10000, // 10 seconds
BACKOFF_MULTIPLIER: 2,
} as const;