FastContext: Training Efficient Repository Explorer for Coding Agents
English summary
FastContext is a system that decouples repository exploration from code solving in LLM coding agents to reduce token waste from irrelevant snippets. It deploys specialized exploration models as a dedicated subagent, issuing parallel tool calls and delivering focused context via concise file paths and line ranges. The approach cuts token consumption by up to 60% and improves resolution rates by up to 5.5% relative to baseline agents.
Chinese summary
FastContext 将仓库探索与代码求解解耦，避免无关代码片段消耗大量 token。它使用专门的探索模型作为子智能体，并行调用工具并仅提供精确的文件路径与行号范围作为上下文。该方法可将 token 消耗降低高达 60%，同时将任务解决率提升最多 5.5%。
Key points
Separates repository exploration from code solving into a dedicated exploration subagent.
将仓库探索从代码求解中分离，由专门的探索子智能体负责。
Specialized exploration models provide concise context in the form of file paths and line ranges instead of full code snippets.
专门的探索模型提供文件路径和行号范围的精确上下文，而非完整代码片段。
Reduces token consumption by up to 60%, significantly improving efficiency.
Token 消耗最高降低 60%，大幅提升效率。
Improves task resolution rates by up to 5.5% over baselines.
相比基线方法，任务解决率最高提升 5.5%。