AudioText documentation¶

AudioText is a CPU-first transcription service that other applications can call over HTTP. It is designed for English, Spanish, and Catalan dictation workflows, with an OpenAI-compatible sync endpoint and an async job API for product flows that need queueing.

Use these docs when you need to install the service, connect an application, operate it on a server, or work with the Python package internals.

Start with the shortest path¶

• Install and run locally
• Deploy with Docker
• Publish the docs to the VPS
• Call the HTTP API
• Operate the service
• Choose a model
• Integrate an application
• Read the Python package reference

What AudioText includes¶

• FastAPI service with /v1/audio/transcriptions and async job endpoints.
• Scoped bearer API tokens with hashed storage.
• Admin login, CSRF protection, and CIDR controls for management routes.
• SQLite persistence for tokens, jobs, settings, model registry entries, audit events, and usage counters.
• CLI commands for setup, tokens, models, settings, workers, health checks, and benchmarks.
• Model runtime caching with unload-all and idle TTL controls.
• Docker, systemd, and launchd deployment templates.

What is intentionally out of scope¶

AudioText does not ship first-party client SDK packages. Client apps should use the HTTP API and OpenAPI-compatible examples in these docs. Real streaming, speaker diarization, GPU deployment profiles, and paid-provider passthrough are later extensions, not V1 defaults.