Tag
Introduces OctoLong, a context engineering pipeline for curating dependency-rich cross-repository code contexts, and OctoLong-Instruct, a suite of long-context open LMs trained on this data. Experiments show that replacing 12% of traditional long-context corpora with OctoLong data yields substantial gains in long-range retrieval, state tracking, repository-level code understanding, and agentic tasks.
archex is a local-first, deterministic tool that builds token-budgeted code context bundles for AI agents, using a full retrieval pipeline (BM25F, local embeddings, cross-encoder reranker, dependency-graph expansion) on your hardware with no API keys or telemetry, outperforming alternatives in recall and efficiency.
WaveScope is an MCP server that applies wavelet transforms to codebases, providing LLMs with multi-resolution structural context to improve code understanding and editing, addressing context rot and structural awareness.
Introduces Ripple, a VS Code extension that analyzes JS/TS projects locally to give AI coding agents context about file dependencies and impact before making edits, addressing the problem of AI agents causing unexpected breakage due to lack of blast radius awareness.
The article argues that giving AI agents access to data through MCP tools (like querying Jira) is not the same as having native structured context like code files. It emphasizes that true understanding requires more than just API access, analogous to having a library card versus having read the books.