mirror of
https://github.com/bilawalsidhu/gods-eye-view.git
synced 2026-11-04 10:24:02 +00:00
Bilawal Sidhu
a19897ad15
- Voice layer manifest (src/voice/layerManifest.js) generates the layer enums, aliases and analyst field hints; a test fails when a catalog layer has no entry. - analyst_query validates fields, operators, units, values and scopes, counts up to 250,000 loaded records per layer, reports complete/partial/unanswered, tells LAYER_OFF / NOT_READY / FEED_UNAVAILABLE apart from zero, refuses mismatched follow-ups and never lets a superseded query replace follow-up memory. Seven more layers expose analyst records. - Tool results carry a code-built say line and a display card; the voice card in the mic control shows captions, plan steps and the result (src/voice/resultDisplay.js is the one adapter). Slow tools narrate their current step out of band; holding Space while the assistant talks cuts it off. Turn spans go to the debug log; caption transcription is metered; the log omits what was said unless GEV_VOICE_LOG_CONTENT=1. - Point and ask: the Space-down pointer snapshot resolves "this", "that one" and "around here" through 'pointer' targets, and "the second one" through referent:n from the last numbered list. - Prompt states each policy once (instructions + tools 48.5k -> 46.7k chars). - Voice benchmark harness (scripts/qa-voice-bench.mjs, scripts/voice-bench/) shares its phrase table and grading with scripts/qa-voice-routing.mjs.
59 lines
2.3 KiB
JavaScript
59 lines
2.3 KiB
JavaScript
/**
|
|
* Cost estimation for benchmark turns (USD). Pure.
|
|
*
|
|
* OpenAI Realtime: the app's own rate table (`src/voice/voiceCost.js`).
|
|
* `gpt-realtime-2.1` has no registry entry, so it resolves to the priciest
|
|
* known table (gpt-realtime-2) — which matches OpenAI's published 2.1 rates
|
|
* as read on 2026-09-23 ($4 / $0.40 cached / $24 per 1M text tokens).
|
|
*
|
|
* Gemini Live: https://ai.google.dev/gemini-api/docs/pricing, read 2026-09-23,
|
|
* gemini-3.8-live paid tier, USD per 1M tokens. Thinking tokens bill as text
|
|
* output. The native-audio Live model rejects TEXT output, so the benchmark
|
|
* requests AUDIO and the spoken confirmation bills at the audio-output rate.
|
|
*/
|
|
import { estimateUsageCostUsd, resolveVoiceModelById } from '../../src/voice/voiceCost.js';
|
|
|
|
export const GEMINI_RATES = Object.freeze({
|
|
'gemini-3.8-live': Object.freeze({
|
|
textInput: 0.75,
|
|
audioInput: 3.0,
|
|
textOutput: 4.5,
|
|
audioOutput: 12.0,
|
|
verifiedOn: '2026-09-23',
|
|
}),
|
|
});
|
|
|
|
/** Cost of one Realtime `response.usage` for `model`. */
|
|
export function openAiUsageCostUsd(model, usage) {
|
|
return estimateUsageCostUsd(usage, resolveVoiceModelById(model).rates);
|
|
}
|
|
|
|
const modalityCount = (details, modality) =>
|
|
(Array.isArray(details) ? details : [])
|
|
.filter((d) => String(d?.modality || '').toUpperCase() === modality)
|
|
.reduce((sum, d) => sum + (Number(d?.tokenCount) || 0), 0);
|
|
|
|
/**
|
|
* Cost of one Gemini Live `usageMetadata`. Unattributed prompt tokens bill
|
|
* as audio input and unattributed response tokens as audio output — the
|
|
* conservative direction for a spend cap.
|
|
*/
|
|
export function geminiUsageCostUsd(model, usage) {
|
|
const rates = GEMINI_RATES[model] || GEMINI_RATES['gemini-3.8-live'];
|
|
if (!usage) return 0;
|
|
const prompt = Number(usage.promptTokenCount) || 0;
|
|
const response = Number(usage.responseTokenCount) || 0;
|
|
const thoughts = Number(usage.thoughtsTokenCount) || 0;
|
|
const textIn = modalityCount(usage.promptTokensDetails, 'TEXT');
|
|
const audioIn = Math.max(0, prompt - textIn);
|
|
const textOut = modalityCount(usage.responseTokensDetails, 'TEXT');
|
|
const audioOut = Math.max(0, response - textOut);
|
|
const usd =
|
|
(textIn * rates.textInput +
|
|
audioIn * rates.audioInput +
|
|
(textOut + thoughts) * rates.textOutput +
|
|
audioOut * rates.audioOutput) /
|
|
1_000_000;
|
|
return Number.isFinite(usd) && usd > 0 ? usd : 0;
|
|
}
|