Agentic Reinforcement Learning (RL) shows promise for complex tasks, but Text-to-SQL remains mostly restricted to single-turn paradigms. A primary bottleneck is the credit assignment problem. In traditional paradigms, rewards are determined solely by the final-turn feedback, which ignores the intermediate process and leads to ambiguous credit evaluation. To address this, we propose Agentic SQL, a framework featuring a universal two-tiered reward mechanism designed to provide effective trajectory-level evaluation and dense step-level signals. First, we introduce Aggregated Trajectory Reward (ATR) to resolve multi-turn credit assignment. Using an asymmetric transition matrix, ATR aggregates process-oriented scores to incentivize continuous improvement. Leveraging Lyapunov stability theory, we prove ATR acts as an energy dissipation operator, guaranteeing a cycle-free policy and monotonic convergence. Second, Column-Set Matching Reward (CSMR) provides immediate step-level rewards to mitigate sparsity. By executing queries at each turn, CSMR converts binary (0/1) feedback into dense [0,1] signals based on partial correctness. Evaluations on BIRD show a 5% gain over binary-reward GRPO. Notably, our approach outperforms SOTA Arctic-Text2SQL-R1-7B on BIRD and Spider 2.0 using identical models, propelling Text-to-SQL toward a robust multi-turn agent paradigm.

Retrieval-Augmented Generation (RAG) enables Large Language Models (LLMs) to access external knowledge during generation, improving factual reliability without parameter retraining. However, existing methods still struggle to retrieve evidence that is both relevant and complete. Standard RAG may miss semantic associations among related chunks. Graph-based RAG addresses this issue by modeling entity relations, but it is often query-agnostic. Question-format RAG further improves query matching by converting facts into Question-Answer (QA) pairs, but it cannot guarantee semantic association and introduces knowledge fragmentation. To address these issues, we propose HolisticRAG, a framework integrating graph-grounded and query-aligned knowledge for holistic knowledge RAG. In the offline stage, HolisticRAG constructs a knowledge representation to preserve semantic associations while reducing query-agnostic retrieval. In the online stage, it applies holistic knowledge clustering to merge all fragmented evidence and adaptively retrieve sufficient supporting facts. Experiments demonstrate the effiectiveness of HolisticRAG.