Every code-search tool makes that pitch; what makes this one usable day to day is that indexing runs in the background, the file watcher keeps it current while you edit, and several codebases can live in one instance at once. Two things to line up before it answers anything: a Qdrant instance on port 6334 and an embedding key in `appsettings.json`.
A .NET 9.0 server that turns local codebases into a semantic index you query in plain language rather than patterns. You point it at a directory, it parses and embeds the code into a Qdrant collection as a background job, and from then on a description — user login validation, database connection pooling, JWT token generation — comes back as ranked passages with file paths and similarity scores. A file watcher keeps the index current as you work, several codebases can be indexed side by side, and indexing tasks are persisted so they survive a restart.
- Semantic search over an indexed codebase from a natural-language description, with the number of passages you want back
- A codebase indexed by absolute path under a friendly name you choose, as a background job with file watching switched on
- Indexing progress and statistics — for one codebase, for one task, or as an overview of every library you have
- An index rebuilt from scratch, for when the code has moved far enough that incremental updates are not enough
- An index and its configuration deleted, behind a second confirming call
- C#-aware parsing that recognises classes, methods and properties rather than chunking the file blindly
.NET 9.0 or newer, a Qdrant vector database reachable on port 6334, and a DashScope API key for embeddings — all three set in `appsettings.json` before the first build. Other embedding providers are supported, including Ollama; DashScope is the one the shipped configuration is written around. The deep structural parsing is specific to C#. The repository carries no licence file.