Give your agents a voice with Smallest AI

New SmallestTools lets an agent generate speech with Smallest AI's Lightning models. Call text_to_speech to turn text into natural audio, or get_voices to see what's on offer. Nothing extra to install beyond agno, since the toolkit runs on Agno's own HTTP client, so it comes down to setting your SMALLEST_API_KEY and going.‍

from agno.agent import Agent
from agno.tools.smallest import SmallestTools

agent = Agent(
    tools=[SmallestTools(voice_id="magnus", model="lightning_v3.1", target_directory="tmp")],
    markdown=True,
)

response = agent.run("Generate a short welcome for a podcast on aviation history.")
# audio saves to tmp/ and is also on response.audio for a later step to use

‍

There are two tiers to work with. lightning_v3.1 is the default and supports cloned voices, while lightning_v3.1_pro opens a premium voice pool with American, British, and Indian accents for broadcast-quality output. Set target_directory and the audio saves itself, or leave it off and pull the bytes off response.audio to pass along however you like. Output can come back as WAV, MP3, or a few telephony formats if you're feeding this into calls.

Check out Agno’s Smallest AI toolkit docs for more.