[FEATURE] Graph input propagation customization #1025
Description
Problem Statement
According to the documentation, there is currently no customization available for how the input is propagated to the next node. Could we have a feature that allows, for example, propagating only the current node’s output as the input to the next node?
Proposed Solution
No response
Use Case
Allowing customization of how input is propagated between nodes offers several benefits:
- Data Isolation & Simplicity: Each node receives only what it needs (e.g., Node B gets only Node A’s output), avoiding confusion and unnecessary dependencies.
- Performance Optimization: Reduces context size, memory usage, and token count.
- Deterministic Execution: Ensures predictable, easier-to-debug workflows.
- Better Context Engineering: Minimizing irrelevant inputs improves efficiency and output accuracy.
Alternatives Solutions
No response
Additional Context
No response