{"query": "list the files worth indexing in a repository by asking git what it tracks, then dropping binaries, lockfiles and minified output", "repo": "code-index", "path": "src/code_index/files.py", "symbol": "discover"} {"query": "cut a source file into overlapping token windows whose edges are pulled onto the nearest function, class or heading start", "repo": "code-index", "path": "src/code_index/chunk.py", "symbol": "chunk_file"} {"query": "keep the pooling strategy and the task prefixes a model was trained with keyed off its filename, and refuse a model that has no entry", "repo": "code-index", "path": "src/code_index/embed.py", "symbol": "PROFILES / profile_for"} {"query": "launch a local inference server and wait for it to answer a health check, because it accepts connections before the weights have loaded", "repo": "code-index", "path": "src/code_index/embed.py", "symbol": "start"} {"query": "grow a memory-mapped float32 matrix on disk by extending the file and rewriting only the row count in its header, instead of rewriting the whole thing", "repo": "code-index", "path": "src/code_index/store.py", "symbol": "append / _write_matrix_header"} {"query": "rank every stored vector against one query vector with a single dense matrix multiply, narrowed first to the rows matching a repository or a path substring", "repo": "code-index", "path": "src/code_index/store.py", "symbol": "search"} {"query": "skip re-doing the expensive work for a file whose contents hash to the same value as the last run", "repo": "code-index", "path": "src/code_index/indexer.py", "symbol": "_reusable_state"} {"query": "take an exclusive file lock so two long-running builds cannot interleave their writes to the same database", "repo": "code-index", "path": "src/code_index/indexer.py", "symbol": "_exclusive_lock"} {"query": "expose the search as a tool an agent can call over a pipe, returning ranked snippets with their line ranges", "repo": "code-index", "path": "src/code_index/server.py", "symbol": "search_code"} {"query": "explain how much memory the model server holds while running and why the batch size is kept small", "repo": "code-index", "path": "README.md"}