Skip to main content
The public SpeechKit module contains the platform-neutral voice kernel. Use it when your application should own its process, configuration, and provider credentials instead of depending on a separate SpeechKit server. The current public release is v0.75.7, published 2026-10-05.

​
Requirements

  • Go 1.26 or newer
  • an application that can provide the selected microphone, audio, or text input
  • provider credentials only when you choose an external provider

​
Add the module

@latest resolves to the newest published release. Record the exact version your build resolved and pin it before shipping: this is beta software, and public APIs, configuration keys, and defaults can change between minor releases. Import only the packages your host needs: pkg/speechkit/dictation for dictation, pkg/speechkit/wakeword for activation, pkg/speechkit/tts for spoken output, pkg/speechkit/companion plus Assist and text-to-speech adapters for one-shot hosts, pkg/speechkit/speaker for speaker-aware apps, pkg/speechkit/meeting for recording sessions, and pkg/speechkit/client for server-connected apps. The root speechkit package holds contracts and value types only. To drive the framework from a config.toml instead of building settings by hand, pkg/speechkit/hostconfig turns a config file into the public mode settings and a starting runtime policy in one call:
A missing file yields the shipped defaults — local only, no cloud keys. The desktop client delegates to the same loader, so an embedder and the reference app read a config.toml identically. A few packages need cgo (Windows audio capture, the ONNX wake word) or an external binary (whisper-server). Each fails closed with a sentinel error when its dependency is absent rather than pretending to work.

​
Other targets

Self-hosted server

A Linux HTTP and WebSocket service over the same kernel, built from public source. It exposes Dictation, Assist, and Voice Agent under one URL.

Agent tools

speechkit-mcp and speechkit-cli let an agent or operator inspect the framework, scaffold starters, validate payloads, and operate a server.

Android modules

Apache-2.0 Gradle modules for server resolution, capture and turn-taking, the wire protocol, and a Compose voice UI, plus an in-repo reference app.

Browser clients

TypeScript packages for the server REST API, the Voice Agent WebSocket protocol, and a framework-neutral voice UI kit, published as source in the repository.

​
Continue with public documentation

These links follow the public repository’s default branch, so they stay valid as the project moves. Pin a tag in the URL when you need the documentation that matches one exact release.