Which Context Engine Finds the Code?

Key findings

1. 01About twice as often as sgrep on held-out issues: a needed file first on 57% of issues, against 30%.
2. 02Almost three times as often as Probe on held-out issues: 57% against 20%.
3. 03More than three times as often as Graphify on our development issues: 83% against 26%.

Abstract

Large language model (LLM) coding agents now fix real issues in real projects, but most of that progress has come in Python and a few other popular languages. C and C++, which still underpin operating systems, browsers, databases and embedded software, have received far less attention, though their large codebases are where agents struggle most to find the code a change needs. In C++, which overload a call reaches, which template is instantiated and which #if branch is compiled all depend on the build, and text search sees none of it. We compare ByteAsk Context, a context engine that understands a codebase the way the project's own build does, with three code search tools, sgrep, Probe and Graphify, giving each the input its own documentation asks for. On real issues from open-source C and C++ projects that were never used in its development, ByteAsk Context put a file that needed changing first about twice as often as sgrep (57% vs 30%) and almost three times as often as Probe (57% vs 20%). On our development issues it did so more than three times as often as Graphify (83% vs 26%).

Cite this paper

Pratyush Saini and Anirudha Kulkarni. Which Context Engine Finds the Code? ByteAsk Research, preprint, November 2026. byteask.ai/research/context-engines-cpp

@misc{saini2026contextengines,
  title        = {Which Context Engine Finds the Code?},
  author       = {Saini, Pratyush and Kulkarni, Anirudha},
  year         = {2026},
  month        = nov,
  institution  = {ByteAsk Research},
  howpublished = {\url{https://byteask.ai/research/context-engines-cpp}},
  note         = {Preprint. Code and data: \url{https://github.com/ByteAsk/context-engines-cpp}}
}