Entire Launches Agentic Search to Give AI Coding Agents the “Why” Behind Code
According to Entire’s benchmark, agents using this new search capability answered 81 of 90 engineering-history questions correctly, compared with 70 of 90 when relying on full Git history and GitHub CLI tools.
Entire has launched Agentic Search, a new API designed to give AI coding agents access to both the code in a repository and the reasoning and development history behind it.
The company said the tool combines semantic search across commits and session transcripts with code search, allowing agents to answer questions about not only what changed in a codebase, but also why those changes were made.
The Left Shift Bureau
According to Entire’s benchmark, agents using this new search capability answered 81 of 90 engineering-history questions correctly, compared with 70 of 90 when relying on full Git history and GitHub CLI tools.
The search-enabled agents also used less than half the tokens, reduced average steps from 14 to seven and cut average cost per question from $0.38 to $0.23.
The benchmark tested nine real engineering-history questions, including questions about why particular design decisions were made. Entire said the most difficult historical question was answered correctly in all 10 runs when search was available, compared with zero correct answers without it.
At the core of Agentic Search are two engines. Firstly, Code Search, which is designed to locate exact code, definitions and references across repositories. Secondly, Semantic Search, which retrieves the context surrounding a change, including commits, transcripts and agent sessions.
Entire said its code-search infrastructure uses an indexing pipeline that responds when developers push changes to a repository. The system updates indexes incrementally where possible and stores them in Amazon S3 while maintaining local caches for faster access. Median query latency is around 100 milliseconds across thousands of indexed repositories.
For semantic search, Entire uses turbopuffer to combine traditional BM25 keyword search with approximate nearest-neighbour semantic search.
Agentic Search is available through the Entire CLI, which can be integrated with coding agents and return structured results through JSON. Developers can search individual repositories, multiple repositories or their entire accessible codebase.
Earlier this year, Entire expanded into India with local data residency for Git repositories to improve Git hosting performance for developers and AI coding agents.
Pritam Bordoloi
The Left Shift Bureau
