MEMCoder: Multi-dimensional Evolving Memory for Private-Library-Oriented Code Generation

2026-04-27

AI Summary

Large language models often underperform on enterprise private codebases due to insufficient understanding of API coordination mechanisms and parameter constraints. This work proposes the first multidimensional dynamic memory framework tailored for private repositories, which jointly models task-level API coordination and API-level parameter constraint knowledge through retrieval-augmented generation (RAG), execution-feedback-driven memory updates, dual-source context injection, and automated closed-loop learning. The framework enables continuous evolution of contextual and constraint-aware representations, achieving an average pass@1 improvement of 16.31% on the NdonnxEval and NumbaEval benchmarks—significantly outperforming existing memory-based continual learning approaches.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Large Language Models (LLMs) excel at general code generation, but their performance drops sharply in enterprise settings that rely on internal private libraries absent from public pre-training corpora. While Retrieval-Augmented Generation (RAG) offers a training-free alternative by providing static API documentation, we find that such documentation typically provides only isolated definitions, leaving a fundamental knowledge gap. Specifically, LLMs struggle with a task-level lack of coordination patterns between APIs and an API-level misunderstanding of parameter constraints and boundary conditions. To address this, we propose MEMCoder, a novel framework that enables LLMs to autonomously accumulate and evolve Usage Guidelines across these two dimensions. MEMCoder introduces a Multi-dimensional Evolving Memory that captures distilled lessons from the model's own problem-solving trajectories. During inference, MEMCoder employs a dual-source retrieval mechanism to inject both static documentation and relevant historical guidelines into the context. The framework operates in an automated closed loop by using objective execution feedback to reflect on successes and failures, resolve knowledge conflicts, and dynamically update memory. Extensive evaluations on the NdonnxEval and NumbaEval benchmarks demonstrate that MEMCoder substantially enhances existing RAG systems, yielding an average absolute pass@1 gain of 16.31%. Furthermore, MEMCoder exhibits vastly superior domain-specific adaptation compared to existing memory-based continual learning methods.

Problem

• private libraries
• code generation
• API coordination
• parameter constraints
• knowledge gap

Innovation

• Multi-dimensional Evolving Memory
• Private-Library-Oriented Code Generation
• Retrieval-Augmented Generation
• Usage Guidelines
• Closed-loop Learning