Switchboard for Voice AI

Help me assess whether any of these Switchboard Voice AI repositories is a useful starting point for my project, either to build a new product or integrate into an existing one.

Start with:
- https://switchboard.audio/cases/voice-ai/ — overview and current availability of the examples.

Inspect these repositories where accessible:
- https://github.com/switchboard-sdk/openai-live-toolkit
- https://github.com/switchboard-sdk/openai-realtime-toolkit
- https://github.com/switchboard-sdk/edgespeech
- https://github.com/switchboard-sdk/hybrid-voice-agent-sample

Use these supporting sources:
- https://switchboard.audio/how-it-works/ — the underlying audio architecture.
- https://docs.switchboard.audio/ — integration and technical documentation; follow relevant platform and extension links.
- https://switchboard.audio/pricing/ — current SDK, licensing, and support information.
- https://synervoz.com/why-synervoz/ — engineering help available for adaptations or larger implementations.

Use what you already know about my project. Establish whether I am building something new or integrating into an existing product. Ask up to five focused questions only if needed about the desired user experience, target platforms/frameworks, offline/local/cloud requirements, existing stack, and practical constraints.

Read the repositories' README files, licences, setup requirements, dependency manifests, and enough relevant source to understand their approach. If you cannot access code or documentation, identify that limitation. Verify availability from current sources; exclude coming-soon projects as runnable starting points.

Compare:
- Actual example platform/framework support.
- Which speech recognition, reasoning, speech synthesis, and audio-processing components run locally or remotely.
- What leaves the device and what works without a network.
- Model/service access, hardware requirements, dependencies, licences, and costs.
- Reusable components versus demonstration-specific code.
- Integration, testing, debugging, and production work still required.

Do not infer behavior solely from names or marketing descriptions. SDK platform support does not establish that a particular example works on those platforms without adaptation. Local speech processing does not make a cloud-dependent agent fully offline.

Return:
1. The best starting point, or "none," with reasons and an optional runner-up.
2. A concise comparison table with evidence links, important limitations, and unknowns.
3. What we could reuse and what we would need to build or change.
4. A simple proposed audio/data flow.
5. The smallest useful next step and how to test its value.

For a new product, propose a small first version based on the best-fitting example. For an existing product, identify a bounded integration that retains useful existing components. Inspect my codebase if it is available, but do not modify it during this assessment.

Identify any concrete compatibility, adaptation, or delivery question where Synervoz could help. If a conversation would be useful, draft a short project brief and three questions; otherwise explain how we can proceed independently. Distinguish documented evidence from inference and do not invent performance, privacy, savings, or readiness guarantees. Do not send anything.