These are some of the core concepts that will be referred to in the documentation going forward, as well as our design philosophy for creating the idTV framework.

Prompts are what allow our agents to execute their given tasks. Below are some examples of base prompts that will be callable through the idTV library, as well as how they feed into the agents.

We separate our prompting library into three specialized classes -- AnchorPrompts for video script generation, NewsPrompts for content analysis, retrieval, and summarization, and VideoPrompts for stylistic implementation. Prompts will be designed to be modular, allowing for easy customization and extension.

Agents are the brains of idTV. They are the models that generate the content, and in some cases orchestrate the subagents that make our broadcast work. Agents can be thought of as a combination of a single model and a single objective. An agent can be given an objective, and may request the usage of other agents through the idTV library. In these cases, results of the functions will be recursively provided to the agent and execution will continue until the objective is complete.

For instance, a NewsProcessor agent may be given a stream of data, and may request the usage of a NewsRetriever agent to fetch relevant information from a database. The NewsProcessor agent will then take the results of the NewsRetriever agent and generate a news script based on the information, then request the usage of a cleanup validation agent to ensure the data is properly formatted for script generation.

No analyst is effective without historical context, and no broadcast is useful without being able to refer to previous interactions and presentations. idTV implements a dual-memory architecture to maintain context and improve decision-making:

Short-Term Memory

• Redis-based caching layer for real-time data

• Stores recent broadcast segments, viewer interactions, and system states

• Configurable TTL for memory management

• Supports pub/sub for real-time updates

Long-Term Memory

• PostgreSQL + pgvector for vector storage

• FAISS for high-performance similarity search

• Stores historical broadcast data, viewer preferences, and content performance metrics

• Implements semantic search for content retrieval