Release Notes: v0.7.0¶

Release Date: 2026-06-15

Highlights¶

Realtime Voice-to-Voice: Native voice-to-voice pipeline with ~100-200ms latency
Dual Pipeline Modes: Choose between "text" (STT→LLM→TTS) or "realtime" (voice-to-voice)
LLM Provider Injection: Optional LLMClient config field for injecting thick providers

New Features¶

Realtime Voice Pipeline¶

The voice gateway now supports native voice-to-voice mode using OpenAI Realtime API or Gemini Live:

import (
    "github.com/plexusone/omni-twilio/omnivoice/gateway"
    coregateway "github.com/plexusone/omnivoice-core/gateway"
    openaiRealtime "github.com/plexusone/omni-openai/omnivoice/realtime"
)

gw, err := gateway.New(gateway.Config{
    AccountSID:  os.Getenv("TWILIO_ACCOUNT_SID"),
    AuthToken:   os.Getenv("TWILIO_AUTH_TOKEN"),
    PhoneNumber: "+15551234567",
    PublicURL:   "https://your-server.com",
    ListenAddr:  ":8080",

    // Enable realtime mode
    Mode: coregateway.PipelineModeRealtime,

    // Configure realtime provider
    RealtimeProvider: openaiRealtime.NewFactory(),
    RealtimeConfig: &coregateway.RealtimeConfig{
        Provider:     "openai",
        APIKey:       os.Getenv("OPENAI_API_KEY"),
        Model:        "gpt-4o-realtime-preview-2024-12-17",
        Voice:        "alloy",
        Instructions: "You are a helpful voice assistant.",
    },
})

Pipeline Modes¶

Mode	Latency	Description
`text`	500-1000ms	Traditional STT → LLM → TTS pipeline
`realtime`	100-200ms	Native voice-to-voice via OpenAI Realtime or Gemini Live

Realtime Providers¶

Provider	Latency	Model
OpenAI Realtime	~100ms	gpt-4o-realtime-preview
Gemini Live	~200ms	gemini-2.0-flash-live

LLM Provider Injection¶

The gateway now supports injecting an LLM provider directly via config, allowing applications to use thick providers (official SDKs) from omnillm:

import (
    "github.com/plexusone/omni-twilio/omnivoice/gateway"
    "github.com/plexusone/omnillm"
)

// Create omnillm client with thick providers
llmClient, _ := omnillm.NewClient(omnillm.ClientConfig{
    Providers: []omnillm.ProviderConfig{
        {Provider: omnillm.ProviderNameAnthropic, APIKey: os.Getenv("ANTHROPIC_API_KEY")},
    },
})

// Inject into gateway
gw, err := gateway.New(gateway.Config{
    // ... other config ...
    LLMClient: llmClient.Provider(), // Inject thick provider
})

If LLMClient is not provided, the gateway uses omnillm-core's thin providers (native HTTP).

New Config Fields¶

Field	Type	Description
`Mode`	`PipelineMode`	Pipeline mode: `"text"` or `"realtime"`
`RealtimeProvider`	`RealtimeProviderFactory`	Factory for creating realtime providers
`RealtimeConfig`	`*RealtimeConfig`	Configuration for realtime sessions
`LLMClient`	`provider.Provider`	Optional pre-configured LLM provider

Installation¶

go get github.com/plexusone/omni-twilio@v0.7.0

Migration Guide¶

From v0.6.0¶

No breaking changes. Existing code using text mode continues to work unchanged.

To enable realtime mode:

Add realtime provider import:

import openaiRealtime "github.com/plexusone/omni-openai/omnivoice/realtime"

Set Mode and realtime configuration:

cfg := gateway.Config{
    // ... existing config ...
    Mode:             coregateway.PipelineModeRealtime,
    RealtimeProvider: openaiRealtime.NewFactory(),
    RealtimeConfig:   &coregateway.RealtimeConfig{...},
}

Changed¶

Replaced omnillm dependency with omnillm-core for minimal dependencies
LLM provider creation now uses omnillm-core registry (thin providers) by default
Applications can inject thick providers via the new LLMClient config field

Dependencies¶

Updated github.com/plexusone/omnivoice-core from v0.11.0 to v0.13.0
Replaced github.com/plexusone/omnillm with github.com/plexusone/omnillm-core v0.17.0
Removed transitive dependencies: omni-anthropic, omni-aws, omni-google, omni-openai

Full Changelog¶

See CHANGELOG.md for the complete list of changes.