Arc Agents to support Voice #47
Description
Arc Agents to support Voice & future roadmap
From a Developers point of view the experience to be like
- Any Agent could be voicified with minimal effort
- Vocification is a System Agent Wrapper that could be plugged into agents needing it ( ie , if the agent developer would like to have the voice capably enabled, the definition and choice lies with the Agent dev to plug in voice within the Agent DSL eg) voice = {in: "gpt:4o STT" , out: "gpt:4o TTS"{
Developer needn't be concerned about all the plumbing. For tests default setting in Arc View would be , for voice input, there is voice output. For text input, there is text output .. In. Arc View the capability to get voice input need to be added
similarly in the future: vision would be added vision= {"gpt:vision"} to accept visual stream to be processed.
The idea that the processing step can product an output can directly be passed on the output channel of the Agent need to be baked in the desing, ..
Example: if the voice model to be used in the future is voice in voice out, the output could b directly sent back as agent response or the output could be channeled on to another component ( module) within the Agent to process it as STT and go through the processing chain ..
The definitions at the top level of the Agent view the agent as a black box and inside the agent there could be multiple chains the developer could flebilxy bring in for other models or steps..
The fundamental principle is: As an agent developer I have all the capabilities available within the framework and to be be realized by the platform , that I don't need to switch back and forth between multiple modules or components.
This is the developer point of view during dev times. And during compilation to a platfrom runtime these capabilities might be provided by the underlaying platform and needn't be the same as how it works in dev setup. But the declarative nature assumes the pipeline is honored , no matter how the implementation is realized.
Output Formatting:
Enabling output formatting for output channels could also be a pluggable feature