opencode-groq-dictate 1.1.2 → 1.1.4

package/README.md

@@ -39,18 +39,23 @@ cursor, as if it had been pasted by hand.
 ## Install
 
 ```bash
-git clone https://github.com/LeonVeganMan/opencode-groq-dictate.git \
-  ~/.config/opencode/plugins/opencode-groq-dictate
+opencode plugin add opencode-groq-dictate
 ```
 
-Local plugins are discovered from `~/.config/opencode/plugins/<name>/`: no entry in
-`opencode.json` is needed. Restart the TUI afterwards — plugins are read at startup and
-there is no hot reload.
+Restart the TUI afterwards — plugins are read at startup and there is no hot reload. On the
+first `alt+v` the plugin asks for the Groq API key (the free tier is enough) and keeps it in
+the opencode plugin storage, never in this repository. If the key is rejected (`401`) it is
+dropped and asked again.
+
+To update: `opencode plugin update`, then restart.
+
+For hacking on it, install from a clone instead — plugins are also discovered from
+`~/.config/opencode/plugins/<name>/` and need no `opencode.json` entry:
 
-The clone above stays the documented install — opencode loads plugins from that folder —
-and the package is also on npm as
-[`opencode-groq-dictate`](https://www.npmjs.com/package/opencode-groq-dictate) for
-directories and registries that index plugins.
+```bash
+git clone https://github.com/LeonVeganMan/opencode-groq-dictate.git \
+  ~/.config/opencode/plugins/opencode-groq-dictate
+```
 
 ## Usage
 
@@ -103,10 +108,13 @@ takes, spend in total.
 
 ## Cost and quota
 
-- Cost is estimated on the audio length at **$0.04/hour** and appended to
+- Cost is estimated on the audio length at the model's rate — **$0.04/hour** for
+  `whisper-large-v3-turbo`, **$0.111/hour** for `whisper-large-v3` — with Groq's 10-second
+  minimum billed length per request. It is appended to
   `~/.cache/groq-dictate-costs.log`, the same file and the same four columns
   (`date time seconds cost`) the older shell script wrote, so a `/dictate-cost` command
-  keeps working unchanged.
+  keeps working unchanged. The model name trails as an optional fifth column, so each
+  line stays attributable while readers of the first four columns ignore it.
 - The **daily** request figure comes from the `x-ratelimit-*` headers of the real
   transcription. No request is spent just to read the quota.
 - The **per-minute** figure is tracked locally: timestamps of the last 60 seconds, cap

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "opencode-groq-dictate",
-  "version": "1.1.2",
+  "version": "1.1.4",
   "description": "Groq dictation in the OpenCode v2 TUI: alt+v records, transcribes with whisper-large-v3-turbo, and pastes the text into the prompt.",
   "private": false,
   "type": "module",

package/src/voice.test.ts

@@ -15,11 +15,17 @@ test("audio length comes from the bytes past the header", () => {
   expect(audioSeconds(44 + 16_000 * 2 * 3)).toBe(3)
 })
 
-test("cost is the hourly rate pro rata", () => {
-  expect(cost(3600)).toBeCloseTo(0.04, 6)
+test("cost follows the model's hourly rate", () => {
+  expect(cost(3600, "whisper-large-v3-turbo")).toBeCloseTo(0.04, 6)
+  expect(cost(3600, "whisper-large-v3")).toBeCloseTo(0.111, 6)
   expect(cost(0)).toBe(0)
 })
 
+test("a request bills at least ten seconds", () => {
+  expect(cost(4, "whisper-large-v3-turbo")).toBeCloseTo((10 / 3600) * 0.04, 9)
+  expect(cost(30, "whisper-large-v3")).toBeCloseTo((30 / 3600) * 0.111, 9)
+})
+
 test("the rpm window drops stamps older than a minute", () => {
   const now = 100_000
   expect(recentRpm([now - 61_000, now - 1_000, now], now)).toEqual([now - 1_000, now])
@@ -41,6 +47,17 @@ test("a config with the wrong types falls back to the defaults", () => {
 
 test("the cost line keeps the shell script's four columns", () => {
   expect(logLine(new Date(2026, 9, 19, 8, 5, 3), 12.34, 0.000137)).toBe("2026-10-19 08:05:03 12.3 0.000137")
+  expect(logLine(new Date(2026, 9, 19, 8, 5, 3), 12.34, 0.000137, "whisper-large-v3")).toBe(
+    "2026-10-19 08:05:03 12.3 0.000137 whisper-large-v3",
+  )
+})
+
+test("the totals ignore the optional model column", () => {
+  const now = new Date(2026, 9, 19, 12, 0, 0)
+  const log = ["2026-10-19 08:05:03 12.3 0.000137", "2026-10-19 09:00:00 30.0 0.000333 whisper-large-v3"].join("\n")
+  const totals = summarize(log, now)
+  expect(totals.today).toBeCloseTo(0.00047, 9)
+  expect(totals.todayCount).toBe(2)
 })
 
 test("the totals only count today's lines, and ignore junk", () => {

package/src/voice.ts

@@ -5,8 +5,13 @@
 export const MODELS = ["whisper-large-v3-turbo", "whisper-large-v3"] as const
 export type Model = (typeof MODELS)[number]
 export const DEFAULT_MODEL: Model = "whisper-large-v3-turbo"
-// Both models bill the same per audio hour; turbo is only faster.
-export const USD_PER_HOUR = 0.04
+// Groq bills per audio hour: large-v3 is ~2.8x turbo, not the same rate.
+export const RATE_PER_HOUR: Record<Model, number> = {
+  "whisper-large-v3-turbo": 0.04,
+  "whisper-large-v3": 0.111,
+}
+// Groq bills a minimum of 10 seconds per request, however short the take.
+export const MIN_BILLED_SECONDS = 10
 export const RPM_LIMIT = 20
 // A forgotten recording would upload silence for minutes; cap it.
 export const MAX_SECONDS = 60
@@ -33,8 +38,9 @@ export function audioSeconds(bytes: number): number {
   return Math.max(0, bytes - WAV_HEADER) / (SAMPLE_RATE * BYTES_PER_SAMPLE)
 }
 
-export function cost(seconds: number): number {
-  return (seconds / 3600) * USD_PER_HOUR
+export function cost(seconds: number, model: Model = DEFAULT_MODEL): number {
+  const billable = seconds > 0 ? Math.max(seconds, MIN_BILLED_SECONDS) : 0
+  return (billable / 3600) * RATE_PER_HOUR[model]
 }
 
 export function recentRpm(stamps: number[], now: number): number[] {
@@ -42,12 +48,14 @@ export function recentRpm(stamps: number[], now: number): number[] {
 }
 
 // Same columns the shell script wrote: `date time seconds cost`, which is what
-// the /dictate-cost command's awk still reads.
-export function logLine(at: Date, seconds: number, usd: number): string {
+// the /dictate-cost command's awk still reads. The model is appended as a fifth
+// column so a rate change stays attributable; older lines simply lack it.
+export function logLine(at: Date, seconds: number, usd: number, model?: Model): string {
   const pad = (n: number) => String(n).padStart(2, "0")
   const date = `${at.getFullYear()}-${pad(at.getMonth() + 1)}-${pad(at.getDate())}`
   const time = `${pad(at.getHours())}:${pad(at.getMinutes())}:${pad(at.getSeconds())}`
-  return `${date} ${time} ${seconds.toFixed(1)} ${usd.toFixed(6)}`
+  const line = `${date} ${time} ${seconds.toFixed(1)} ${usd.toFixed(6)}`
+  return model ? `${line} ${model}` : line
 }
 
 export type Totals = { today: number; total: number; todayCount: number }

package/tui.tsx

@@ -17,6 +17,7 @@ import {
   recentRpm,
   startTicker,
   summarize,
+  type Model,
 } from "./src/voice"
 
 const WAV = "/tmp/opencode-groq-dictate.wav"
@@ -119,10 +120,10 @@ export default Plugin.define({
     }
 
     // Cost bookkeeping: the log line is best effort, the totals are in memory.
-    async function bookkeep(seconds: number) {
-      const usd = cost(seconds)
+    async function bookkeep(seconds: number, model: Model) {
+      const usd = cost(seconds, model)
       try {
-        await appendFile(COST_LOG, logLine(new Date(), seconds, usd) + "\n")
+        await appendFile(COST_LOG, logLine(new Date(), seconds, usd, model) + "\n")
       } catch (error) {
         context.ui.toast.show({ message: `Cost log: ${String(error)}`, variant: "warning" })
       }
@@ -155,7 +156,7 @@ export default Plugin.define({
       // a copy of everything dictated on disk.
       await rm(WAV, { force: true }).catch(() => {})
 
-      const usd = await bookkeep(seconds)
+      const usd = await bookkeep(seconds, model)
       context.ui.toast.show({ message: `Dictated ${seconds.toFixed(1)}s · $${usd.toFixed(4)}`, variant: "success" })
     }