July 2026 Summaries

Agora has introduced a major update to its Console, featuring an AI assistant called Concierge that simplifies the process of creating or modifying voice agents through plain-language requests. Users can request the creation of an agent, review the proposal, and confirm changes, all within the same interface, resulting in a more efficient workflow without the need to switch tools. The Console's new features include real-time voice processing, customizable agent parameters, and the ability to test calls directly in the browser before deployment. Additionally, the update enhances accessibility with light and dark themes and keyboard navigation improvements, while also consolidating project management tools for a streamlined user experience. Agora offers free trials to attract new users and is promoting its solutions in various sectors, including an upcoming webinar on the impact of AI and AR/VR in telehealth.

OpenAI's GPT-Live, launched on July 8, 2026, aims to enhance AI-powered voice interactions by introducing a full-duplex mode that listens and responds simultaneously, unlike its predecessors which operated on a turn-based system. Agora Media Lab conducted tests revealing GPT-Live's ability to handle interruptions and background speech more effectively than earlier models, although it occasionally misattributed background voices as user input in noisy environments. The new model showed improved consistency in response times, reducing jitter significantly, which contributes to a perception of naturalness in conversations. GPT-Live also demonstrated resilience to network issues, outperforming previous models even under 10% packet loss. However, achieving a seamless user experience requires synchronization across the model, client, network, and playback, which OpenAI manages by controlling the entire stack, unlike most teams that rely on a model API over the public internet. The findings highlight the importance of consistent timing and network conditions in delivering natural full-duplex voice interactions.

Agora has introduced a new command-line interface (CLI) designed to streamline the development process for developers and AI coding agents working with its platform. This CLI moves the setup path into the terminal, allowing users to write environment files, scaffold starter repositories, and check environment readiness without leaving their current workflow. It features a JSON output mode, enabling structured responses that coding agents can parse and act upon, facilitating automated workflows. The CLI includes commands for project creation, authentication, diagnostics, and more, with templates available for Next.js, Python, and Go projects. It aims to reduce setup complexities while maintaining the alignment of quickstart projects, credentials, and project bindings. The CLI supports both human and agent use cases, enhancing the integration of Agora's real-time conversational AI tools into development environments.

Agora has introduced the Agora Agents SDK, a typed Software Development Kit designed for building voice agents using Agora's Conversational AI Engine, available in Python, Node.js, and Go. This SDK facilitates the creation of voice agents by providing typed builders, built-in authentication, token generation, and session management, which sit atop the existing REST API. Developers can easily configure voice agents with speech-to-text, language models, and text-to-speech functionalities, while dynamic token generation and session lifecycle management streamline the process. The SDK does not restrict developers to specific model stacks, allowing them to bring their own models or use Agora-managed ones, while the real-time infrastructure handles media paths and channel lifecycles. The REST API remains supported, and the SDK simplifies complex code, making it more accessible for teams that prefer not to manage extensive backend logic. This initiative by Agora aims to reduce the need for intricate glue code, aligning the setup process closely with the agent's intended functionalities, and feedback from users is encouraged to refine future iterations.