an open-source alternative to the dots, bots and muses of the world, run an office of claude code/codex like agents on your laptop, sandboxes or anywhere, uses your existing subscriptions
HTML agent-harnessagent-orchestrationagentsai-agentsautonomous-agentsclaude-codecodexcoding-agentsdesktop-appdeveloper-toolselectrongemini-cliharnessharness-engineeringlocal-firstmemorymulti-agentopencodeorchestrationtypescript
/**
* Realtime Michael — audio token pricing for the OpenAI gpt-realtime-2 voice loop
* (card rt-9, cost-guard). Shared so BOTH the main cost module (src/main/
* realtimeCost.ts) and the renderer cost store (src/renderer/src/realtime/
* costStore.ts) price from one source. Pure + dependency-free (no electron, no
* main-only imports) so it's safe to import from either process.
*
* Pricing (locked plan / board): gpt-realtime-2 AUDIO tokens bill at $32 per 1M
* input and $64 per 1M output. A voice turn is dominated by audio; text tokens
* (instructions, tool args) and cached tokens are cheaper. We deliberately price
* ALL input/output tokens at the audio rate: for a cost GUARD a conservative
* UPPER bound is the safe choice (warn early rather than under-count), and it uses
* only the authoritative audio numbers rather than guessing text/cache rates. The
* raw token counts are surfaced alongside the dollar figure so it stays auditable.
*/
/**
* The realtime voice model Talk connects to. Lives HERE, in shared, because it is
* named in user-facing Settings copy as well as used by the main-process mint —
* a model string duplicated across a process boundary is one that eventually
* disagrees with itself, and the half users read is the half nobody notices is
* stale. `src/main/realtime.ts` re-exports this as its own REALTIME_MODEL.
*/
export const REALTIME_MODEL = 'gpt-realtime-2.1';
/** USD per 1,000,000 audio tokens (gpt-realtime-2). */
export const REALTIME_AUDIO_INPUT_PER_MTOK = 32;
export const REALTIME_AUDIO_OUTPUT_PER_MTOK = 64;
/**
* A usage delta as reported by the realtime session. Accepts both the SDK's
* camelCase (`inputTokens`) and the raw realtime API's snake_case
* (`input_tokens`) so the caller can forward whatever shape it has.
*/
export interface RealtimeUsage {
inputTokens?: number;
outputTokens?: number;
totalTokens?: number;
input_tokens?: number;
output_tokens?: number;
total_tokens?: number;
}
/** Pull input/output token counts out of either casing; missing ⇒ 0. */
export function normalizeRealtimeUsage(u: RealtimeUsage | null | undefined): {
inputTokens: number;
outputTokens: number;
} {
const n = (v: number | undefined): number => (typeof v === 'number' && isFinite(v) && v > 0 ? v : 0);
return {
inputTokens: n(u?.inputTokens) || n(u?.input_tokens),
outputTokens: n(u?.outputTokens) || n(u?.output_tokens)
};
}
/**
* USD cost for ONE realtime usage delta. Conservative upper bound: every input
* token is priced at the audio-input rate and every output token at the
* audio-output rate (see file header).
*/
export function computeRealtimeUsd(u: RealtimeUsage | null | undefined): number {
const { inputTokens, outputTokens } = normalizeRealtimeUsage(u);
return (
(inputTokens / 1_000_000) * REALTIME_AUDIO_INPUT_PER_MTOK +
(outputTokens / 1_000_000) * REALTIME_AUDIO_OUTPUT_PER_MTOK
);
}
/** Compact USD formatter: <$1 shows cents (e.g. $0.42), else 2dp (e.g. $12.30). */
export function formatUsd(n: number): string {
if (!isFinite(n) || n <= 0) return '$0.00';
if (n < 0.01) return '<$0.01';
return `$${n.toFixed(2)}`;
}