AI code intelligence for C# monorepos
GraphSlice turns your C# solution into a Code Property Graph and exposes only the context an LLM actually needs to answer a question.
Built on Roslyn Delivered over MCP Runs on-premise
A modern C# solution holds millions of relationships between types, methods, namespaces, dependencies and data flows. Dumping files into a model, or hoping a vector search retrieves the right ones, is slow, expensive and rarely produces a reliable answer.
GraphSlice builds a semantic graph of your code and serves targeted graph slices over the Model Context Protocol. Instead of searching text, your AI assistant queries how the software actually fits together.
By targeting only the exact dependency paths a prompt needs, you get drastically reduced tokens per query and higher precision in code generation, eliminating context drift and keeping costs flat even as your repository grows.
How it works
GraphSlice uses Roslyn to fully compile and understand your C# solution: the same semantic model the compiler uses, not a text approximation.
Compiler output becomes a Code Property Graph. Every symbol is connected to the ones it touches, so the graph reflects how your software behaves, not where text happens to appear.
Rather than expose the whole graph, GraphSlice extracts only the region a question needs.
Slices are delivered through MCP. Any MCP-compatible client can request semantic code context, without indexing or uploading your repository itself.
Why GraphSlice
GraphSlice reads C# semantics through Roslyn instead of generic Tree-sitter parsers. Queries operate on real symbols and relationships, including external packages and BCL boundaries. No filenames. No grep. No embeddings to drift out of date.
Only the relevant slice reaches the model. Lower token usage, higher accuracy and faster responses, even on solutions with thousands of types.
Designed from the ground up for LLM workflows. Bring your own MCP client: the graph does the reasoning about structure so the model can reason about intent.
Validated against large, real-world C# codebases, not toy repositories.
Example
Explain how a customer order reaches billing.
Produces an explanation grounded in architecture-aware context, instead of guessing from unrelated files.
Sovereignty
GraphSlice is built in Belgium by an independent European team for organizations that refuse to compromise on data sovereignty. Your source code is one of your most valuable assets. It should never have to leave your infrastructure to benefit from AI.
Run GraphSlice on-premises, in your private cloud, or in an air-gapped environment. Your repositories never leave your infrastructure, because neither should your intellectual property.
Choose the AI you trust. Connect GraphSlice to your own LLM or MCP client, whether it's self-hosted or based on open-weight models. There's no vendor lock-in, no mandatory cloud service, and no code shared with any model provider you didn't explicitly choose.
Privacy by default. Sovereignty by design. GraphSlice is built for European organizations with strict security, compliance, and data sovereignty requirements, where protecting intellectual property is non-negotiable.
Alpha
GraphSlice is open to a small group of early teams building large C# applications who want better AI-assisted development. Tell us a little about your codebase and we'll be in touch.