A fixed LLM workflow for more careful AI companion replies
Many attempts at emotionally aware AI rely on one instruction that tells the model to sound warm and supportive. That can produce shallow reassurance. This breaks into several LLM steps instead of trying to do in one call.
It is a workflow, not an agent. The order is fixed in code, and the model does not decide what to do next. The flow is router, analyzers, composer, and verifier.
Only the router uses an LLM decision to choose the path for the incoming message. There are no that change outside systems. Under Anthropic’s distinction, an agent drives its own next steps, while this setup places inside a predefined path.
Key points
- A single warm-and-supportive prompt can lead to shallow emotional replies.
- The design splits into multiple LLM steps.
- The pipeline is router, analyzers, composer, and verifier.
- The model does not control the full execution order; code does.
- There are no that change external state, so this is a workflow rather than an agent.