gradbot 0.1.3

Python bindings for gradbot voice AI library

gradbot

Python bindings for the gradbot voice AI library — real-time speech-to-speech with tool calling.

Installation

pip install gradbot

Environment Variables

Variable	Required	Description
GRADIUM_API_KEY	Yes	API key for Gradium STT/TTS services
LLM_API_KEY	Yes	API key for OpenAI-compatible LLM
LLM_BASE_URL	No	LLM API base URL (defaults to OpenAI)
LLM_MODEL	No	LLM model name (auto-detected if only one available)
GRADIUM_BASE_URL	No	Base URL for Gradium services

Quick Start

import asyncio
import gradbot

async def main():
    input_handle, output_handle = await gradbot.run(
        session_config=gradbot.SessionConfig(
            voice_id="YTpq7expH9539ERJ",
            instructions="You are a helpful assistant.",
            language=gradbot.Lang.En,
        ),
        input_format=gradbot.AudioFormat.OggOpus,
        output_format=gradbot.AudioFormat.OggOpus,
    )

    while True:
        msg = await output_handle.receive()
        if msg is None:
            break
        if msg.msg_type == "audio":
            play(msg.data)  # bytes
        elif msg.msg_type == "tool_call":
            result = handle(msg.tool_call.tool_name, msg.tool_call.args_json)
            await msg.tool_call_handle.send(result)

asyncio.run(main())

Remote Mode

Connect to a gradbot_server instead of running STT/LLM/TTS locally:

input_handle, output_handle = await gradbot.run(
    gradbot_url="wss://your-server.com/ws",
    gradbot_api_key="grd_...",
    session_config=config,
    input_format=gradbot.AudioFormat.OggOpus,
    output_format=gradbot.AudioFormat.OggOpus,
)

When gradbot_url is set, all other client params are ignored — the server handles everything.

FastAPI Integration

The gradbot.fastapi module provides a WebSocket handler and route setup for building voice demos.

from fastapi import FastAPI, WebSocket
from gradbot.fastapi import websocket_chat_handler, setup_demo_routes

app = FastAPI()

setup_demo_routes(app, static_dir="static", voices=True)

@app.websocket("/ws")
async def ws(websocket: WebSocket):
    await websocket_chat_handler(
        websocket,
        on_start=lambda msg: gradbot.SessionConfig(
            instructions="You are a helpful assistant.",
        ),
    )

setup_demo_routes registers /api/audio-config, serves your static files, and automatically serves the bundled JS audio processor at /static/js/.

WebSocket Protocol

Direction	Format	Description
Client → Server	JSON {"type": "start", ...}	Begin session
Client → Server	Binary	Audio data
Client → Server	JSON {"type": "config", ...}	Reconfigure mid-session
Client → Server	JSON {"type": "stop"}	End session
Server → Client	JSON	Transcripts, events, audio timing
Server → Client	Binary	Audio data

API Reference

Functions

• run(...) — Create clients and start a session. Returns (SessionInputHandle, SessionOutputHandle).
• create_clients(...) — Create reusable GradbotClients for multiple sessions.
• flagship_voices() — List all available voices.
• flagship_voice(name) — Look up a voice by name (case-insensitive).
• voices_json() — All voices as JSON-serializable dicts.
• voice_switching_tools() — ToolDef list for switch_to_{name} tools.
• resolve_voice_from_tool(tool_name) — Resolve a switch_to_* tool name to a FlagshipVoice.
• init_logging() — Initialize debug logging.

Enums

Enum	Values
Lang	En, Fr, Es, De, Pt
Gender	Masculine, Feminine
Country	Us, Gb, Fr, De, Mx, Es, Br
AudioFormat	OggOpus, Pcm (24kHz in / 48kHz out), Ulaw (G.711 mu-law)

Classes

• SessionConfig — voice_id, instructions, language, assistant_speaks_first, silence_timeout_s, tools
• ToolDef — name, description, parameters_json
• SessionInputHandle — send_audio(bytes), send_config(SessionConfig), close()
• SessionOutputHandle — receive() -> MsgOut | None
• MsgOut — msg_type is one of "audio", "tts_text", "stt_text", "event", "tool_call"
• ToolCallInfo — call_id, tool_name, args_json
• ToolCallHandlePy — send(result_json), send_error(error_message)

Examples

See demos/ for complete examples including tool calling, voice switching, and WebSocket frontends.