Answering complex factoid questions often requires synthesizing information across diverse knowledge sources. Traditional methods to question-answering (QA) over multiple knowledge bases (KBs) involving merging all KBs into a single graph through entity alignment, but this approach overlooks diverse link types (full/partial) between KBs and compromises the expressive flexibility inherent in distinct KBs and their ontologies. We propose Multi-KB-QA, a new task and a novel framework that synergistically integrates multiple KBs without requiring full alignment. Our key innovation treats unalignable entities representing different aspects of abstract concepts as equivalent when referenced in questions, enabling unified encoding of cross-KB relationships for effective answer scoring. To evaluate this paradigm, we introduce a benchmark featuring diverse inter-KB relations and complex query types. Experiments show our method significantly outperforms conventional KB-QA systems in multi-source reasoning, highlighting the value of preserving structured information’s richness while enabling collaborative KB reasoning. This work advances QA capabilities for real-world scenarios where answers emerge from interconnected but heterogeneously structured knowledge.