CastReader Voice APICASTREADER FOR CODEX · 0.1
Build speech into your application, or turn a paragraph into a playable audio file. Keep the code, the audio and a clear record of what it cost.
Git marketplace release. OpenAI public-directory approval is a separate process.
A real generated file
Hello. CastReader turns your ideas into playable audio.
Generated September 15, 2026 in the China processing region. File decoding and saved-file reuse verified. One first-party sample, not a quality benchmark.
Inspect file receipt →01 · BUILD
“Add text-to-speech to my app and run it locally.”
Codex checks the current API, implements backend credentials and queued jobs, adds playback, and verifies the result in your chosen stack. A complete Node.js audio-card starter is included.
$build-voice-app02 · GENERATE
“Turn this paragraph into an MP3. Show the cost.”
Codex estimates the actual input, saves the request before submitting it, downloads a successful result and returns playable audio with its charge. Resume the same folder after interruption.
$generate-audiocodex plugin marketplace add vinxu/castreader-codex
codex plugin add castreader@castreaderYou can build and inspect the local app before configuring a key. Live synthesis requires your activated account.
Current API rate: $8 per million normalized characters, before available trial credit. The plugin checks live estimates; waiting and downloading retained results do not create a second generation charge.
node scripts/voice.mjs resume --output ./audio-run --wait-ms 45000An estimate is not a reserved quote. Failed, expired and cancelled jobs are reported without automatic replacements. Keep private state.json and save audio within the current 24-hour retention window.
Queued MP3/WAV, 500 characters per job, explicit language selection. No streaming, realtime calls or automatic translation. English timestamps are optional and may be unavailable.
Two skills: build a working speech feature in your app, or generate an MP3/WAV with a usage receipt. The package includes a runnable local audio-card app and a dependency-free audio command.
No. API trial and wallet credit are separate from the reading app subscription. The plugin estimates actual text before submitting a job and reports its final charge.
Keep the output folder and ask Codex to resume it. The saved request body, region and idempotency key identify the original task. Valid local audio is reused; missing files are downloaded from the same retained result.
The Git marketplace package can be installed using the commands on this page. OpenAI directory review and publication are separate; this page does not claim directory approval or automatic recommendation.
Version 0.1 uses local skills and a server-side API key. Hosted MCP, OAuth account linking and reading-library actions are planned follow-up capabilities. They are not included in this package.
Report installation issues with your Node/Codex versions and sanitized error code. Keep API keys and private job state out of public issues.