Skip to content

Release Notes: v0.13.0

Release Date: 2026-07-05

Overview

OmniAgent v0.13.0 introduces realtime voice mode for LiveKit agents, enabling ultra-low latency (~100-300ms) native voice-to-voice conversations. This release also adds Deepgram as a realtime voice provider, LiveKit voice agent commands, and role manager integration for agent personas.

Highlights

  • Realtime Voice Mode - LiveKit agents now support native voice-to-voice with ~100-300ms latency
  • Deepgram Agent API - New realtime provider via omni-deepgram for voice conversations
  • LiveKit Voice Agents - New cmd/livekit-agent and cmd/livekit-agent-facilitator commands
  • Role Manager - Agent personas with skills, workflows, and policy enforcement

New Features

Realtime Voice Mode for LiveKit Agents

The livekit-agent-facilitator now supports realtime voice mode, providing native voice-to-voice conversations with significantly lower latency than the traditional STT→LLM→TTS pipeline.

# Set credentials
export LIVEKIT_URL="wss://your-project.livekit.cloud"
export LIVEKIT_API_KEY="your-api-key"
export LIVEKIT_API_SECRET="your-api-secret"

# Realtime mode with Deepgram
export REALTIME_PROVIDER="deepgram"
export DEEPGRAM_API_KEY="your-deepgram-key"

# Run the facilitator
go run ./cmd/livekit-agent-facilitator

Deepgram Realtime Provider

Deepgram's Agent API is now available as a realtime voice provider:

Provider Package Latency Voices
OpenAI Realtime omni-openai/omnivoice/realtime ~100ms 11 voices
Gemini Live omni-google/omnivoice ~200ms 5 voices
Deepgram Agent omni-deepgram/omnivoice/realtime ~200ms 10+ voices

Configuration:

omniagent voice serve \
  --provider twilio \
  --realtime deepgram \
  --realtime-voice aura-2-thalia-en \
  --public-url https://your-server.com

Available Deepgram voices: aura-2-thalia-en, aura-2-andromeda-en, aura-2-arcas-en, aura-2-luna-en, aura-2-orion-en, aura-2-perseus-en, aura-2-stella-en, aura-2-zeus-en

Voice Mode Comparison

Mode Latency Configuration Best For
Realtime 100-300ms Single provider Low latency conversations
Traditional 500-1500ms Separate STT/LLM/TTS Custom voice selection

LiveKit Voice Agent Commands

New commands for running voice agents in LiveKit rooms:

Command Description
cmd/livekit-agent Generic voice agent with web search
cmd/livekit-agent-facilitator Meeting facilitator with Meeting PM role 
# Set credentials
export LIVEKIT_URL="wss://your-project.livekit.cloud"
export LIVEKIT_API_KEY="your-api-key"
export LIVEKIT_API_SECRET="your-api-secret"

# Run generic agent
go run ./cmd/livekit-agent

# Run facilitator with role
go run ./cmd/livekit-agent-facilitator

Role Manager Integration

The agent/roles package provides role management for agent personas:

import (
    "github.com/plexusone/omniagent/agent/roles"
    facilitator "github.com/plexusone/omnirole-facilitator"
)

// Create role with configuration
pmRole := facilitator.New(facilitator.Config{
    DefaultConfluenceSpace: "TEAM",
    EnableActionTracking:   true,
})

// Create role manager
mgr, _ := roles.NewManager(pmRole, meetingSkill, googleSkill)
mgr.Init(ctx)

// Access role capabilities
prompt, _ := mgr.SystemPrompt(ctx)
workflows := mgr.Workflows()

New Environment Variables

Variable Description Default
REALTIME_PROVIDER Realtime provider: openai, gemini, deepgram -
REALTIME_VOICE Voice for realtime API Provider default

Configuration Changes

Updated Fields

Field Change
voice.realtime.provider Added deepgram option

Realtime Configuration

voice:
  enabled: true
  realtime:
    provider: deepgram       # openai, gemini, or deepgram
    voice: aura-2-thalia-en  # Provider-specific voice

Dependency Updates

Package Version Change
github.com/plexusone/omni-deepgram v0.7.0 Added realtime provider
github.com/plexusone/omnivoice v0.12.1 → v0.13.0 Deepgram realtime factory
github.com/plexusone/omni-livekit v0.1.0 LiveKit voice transport (new)
github.com/plexusone/omnillm v0.16.1 → v0.17.0 LLM provider updates
github.com/plexusone/omnillm-core v0.18.0 Core LLM types
github.com/plexusone/omnimeet-core v0.1.0 Meeting abstractions (new)

Bug Fixes

  • Fixed G115 gosec warnings in PCM audio processing code
  • Proper nolint annotations for intentional integer conversions

Upgrade Guide

From v0.12.0

  1. Update your dependency:
go get github.com/plexusone/omniagent@v0.13.0
go mod tidy
  1. (Optional) Enable realtime voice mode:
export REALTIME_PROVIDER=deepgram
export DEEPGRAM_API_KEY=your-key
  1. Run LiveKit agent with realtime:
go run ./cmd/livekit-agent-facilitator

Documentation

New guides:

Updated guides:

Full Changelog

See CHANGELOG.md for the complete list of changes.