wtf is synapse?

Synapse gives the Claude AI assistant knowledge of your own codebase without sending that code to the cloud. Normally when you ask an AI coding assistant for help, you have to explain your project structure, paste in relevant files, and describe your conventions every time. Synapse solves this by watching your project folders, breaking the files into chunks, and turning them into a searchable local index, so Claude can pull up the exact code it needs on its own. You install it with a single pip command, then run a setup command pointing at your project folder. After adding one line to your Claude configuration, Claude connects to Synapse through a protocol called MCP, which lets AI tools call external programs. From then on, whenever you ask Claude a question about your code, it can automatically search the index and pull in the relevant snippets before answering. Everything happens on your own machine by default. Synapse uses a free, local model to turn your code into searchable data, and the only network request is a one time download of that model, around 400 megabytes. There is an optional setting to use OpenAI instead for slightly better search quality, but that sends your code to OpenAI's servers, so the README recommends the local option for private or sensitive projects. A file watcher keeps the index updated automatically as you save changes. Synapse supports many common file types including Python, JavaScript, TypeScript, Go, Rust, Java, and configuration formats like Docker and Makefiles. It works with Claude Code and Cursor today, with Claude Desktop chat support planned for a future release, and it is designed to work with any tool that speaks the MCP protocol rather than being locked to one app. The project is MIT licensed and open to contributions, with a roadmap that includes indexing git history and building out a search interface in the terminal.