Yo everyone — I want to revisit my earlier post because GitHub just dropped something that's going to change Copilot CLI from a useful tool into an absolute game-changer! Right now the CLI largely relies on semantic search plus grep across your codebase before sending context to the model, which is basically text matching and it works... but poorly. It means irrelevant code gets dragged along for the ride because of fuzzy matches, and worse — it can't resolve real dependencies between modules at all since Grep has no concept of imports or types. That's like trying to understand a sentence by only looking at nearby words instead of following pronouns across paragraphs; you end up with answers that look right but are fundamentally disconnected from how your code actually runs.
But the new update adds full language server protocol (LSP) support and this is the massive upgrade I was hoping for! Instead of blind Grep, Copilot now builds an in-memory symbol graph of your repo before answering any CLI query — it uses actual LSP servers to resolve imports, function definitions across module boundaries, type hierarchies, and even cross-file dependencies. So when you ask a question about a type defined two packages away the tool can follow the import chain through real language server symbols rather than guessing based on string similarity. It drastically reduces irrelevant context sent to the LLM while including exactly what's needed, which means answers are sharper and far less prone to hallucinations from unrelated code snippets nearby. I am absolutely stoked about this because it moves Copilot closer to understanding my project as a real program rather than just a pile of text files — that's a huge win for anyone working in large monorepos or multi-module projects where Grep always falls apart!
Source: https://github.blog/ai-and-ml/github-copilot/give-github-copilot-cli-real-code-intelligence-with-language-servers/
But the new update adds full language server protocol (LSP) support and this is the massive upgrade I was hoping for! Instead of blind Grep, Copilot now builds an in-memory symbol graph of your repo before answering any CLI query — it uses actual LSP servers to resolve imports, function definitions across module boundaries, type hierarchies, and even cross-file dependencies. So when you ask a question about a type defined two packages away the tool can follow the import chain through real language server symbols rather than guessing based on string similarity. It drastically reduces irrelevant context sent to the LLM while including exactly what's needed, which means answers are sharper and far less prone to hallucinations from unrelated code snippets nearby. I am absolutely stoked about this because it moves Copilot closer to understanding my project as a real program rather than just a pile of text files — that's a huge win for anyone working in large monorepos or multi-module projects where Grep always falls apart!
Source: https://github.blog/ai-and-ml/github-copilot/give-github-copilot-cli-real-code-intelligence-with-language-servers/