build a voice agent with livekit

building on livekit agents? use livekit-plugins-rumik, our official livekit tts plugin. it drops rumik straight into an AgentSession with streaming audio and interruption handling already wired up.

livekit-plugins-rumik-ai is deprecated. use livekit-plugins-rumik instead. see migrating from livekit-plugins-rumik-ai.

pip install livekit-plugins-rumik

set your key from the dashboard:

export RUMIK_API_KEY="rk_live_•••••••••"

add it to an agent

the TTS class plugs into a livekit AgentSession next to your stt and llm:

from livekit.agents import AgentSession
from livekit.plugins import rumik

# muga: expressive, tone-tagged hinglish
session = AgentSession(
    stt=...,   # your speech-to-text plugin
    llm=...,   # your llm plugin
    tts=rumik.TTS(model="muga"),
)

steer mulberry with a natural-language description, or pin a preset speaker:

# mulberry: description-driven voice
tts = rumik.TTS(
    model="mulberry",
    description="a female 30s hindi voice, warm timbre, conversational pacing, like a podcast host",
)

# or pin a named studio voice. description still goes with it.
tts = rumik.TTS(
    model="mulberry",
    description="a female 30s hindi voice, warm timbre, conversational pacing, like a podcast host",
    speaker="ira",
)

constructor options

argument	applies to	notes
`model`	both	`"muga"` or `"mulberry"`. default `"muga"`.
`api_key`	both	defaults to the `RUMIK_API_KEY` environment variable.
`base_url`	both	defaults to `https://silk-api.rumik.ai`.
`full_response_aggregation`	both	buffer the full reply before synthesis. default `True` for muga, `False` for mulberry.
`tone`	muga	fallback tone when the input text has no `[tone]` marker.
`description`	mulberry	required. natural-language voice description.
`speaker`	mulberry	optional. a named voice, e.g. `"ira"`. send `description` too. see preset voices.

it also accepts the shared sampling params (temperature, top_p, top_k, repetition_penalty, max_new_tokens).

muga aggregates the full tagged reply before speaking (full_response_aggregation is True by default), so it never tries to synthesize a half-tagged sentence. see prompting muga for why.

which model?

muga for short, expressive reactions you steer with [tone] tags. see prompting muga.
mulberry for low-latency conversational agents you steer with a description. see prompting mulberry.

prefer a full walkthrough? the livekit cookbook builds a working voice agent from scratch. or hand your coding agent the rumik tts skill so it wires all of this up on the first try.

migrating from livekit-plugins-rumik-ai

livekit-plugins-rumik continues livekit-plugins-rumik-ai from 0.2.3 on. the TTS class and its options are the same; only the package and module names changed. swap the package in your environment, and in your requirements.txt or pyproject.toml:

pip uninstall livekit-plugins-rumik-ai
pip install "livekit-plugins-rumik>=0.2.3"

then change the import. importing the new module under the old name leaves the rest of your code as it is:

# before
from livekit.plugins import rumik_ai

# after
from livekit.plugins import rumik as rumik_ai

rebuild and redeploy your agent. if you filter logs by logger name, the plugin now logs as livekit.plugins.rumik. beyond the rename, 0.2.3 keeps the connection alive between replies, so a pause no longer ends in Connection error., and it opens replacement connections in the background. when a connection does fail, the error now names the cause. see the changelog.

documentation

sdk

integrations

cookbook

api reference

build a voice agent with livekit

add it to an agent

constructor options

which model?

migrating from livekit-plugins-rumik-ai

​add it to an agent

​constructor options

​which model?

​migrating from livekit-plugins-rumik-ai

add it to an agent

constructor options

which model?

migrating from livekit-plugins-rumik-ai