How to Use the AI Chat Panel in Cursor (Cmd/Ctrl+L)

Cmd+L (Ctrl+L on Windows/Linux) opens Cursor's AI Chat sidebar panel, providing a fuller conversational interface with the AI for questions, larger or multi-file requests, and iterative discussion beyond what a single scoped inline edit comfortably handles. **When Chat is the better choice over inline edit**: open-ended questions about how existing code works, requests spanning multiple files or requiring broader codebase context, or situations where you want to discuss and refine an approach conversationally before committing to a specific implementation are generally better suited to Chat than the more immediate, scoped nature of inline editing. **Grounding the conversation with specific code**: Cmd/Ctrl+Shift+L pins a highlighted block into the message you're about to send, so the model reads the actual lines you're asking about instead of guessing at them from a text description alone — a meaningful accuracy difference when the question hinges on an exact variable name, return type, or edge case buried in the implementation. The @ trigger inside the input box goes a step further by opening a searchable picker for files, symbols, or connected docs, giving you a way to point the AI at something you haven't already selected in the editor rather than being limited to whatever text happens to be highlighted. **Escalating to Agent mode**: toggling Agent mode (Cmd/Ctrl+I) from within the same panel changes the interaction fundamentally — instead of waiting for your explicit go-ahead after each proposed step, the AI can search the codebase, run terminal commands, and carry out a multi-step plan spanning several files on its own initiative, useful for larger, well-defined tasks where you'd rather review the end result than approve every intermediate step individually. **How responses relate to your actual files**: Chat responses may include proposed code changes or full snippets, but unlike inline edit's direct in-place application, applying a Chat-proposed change to your actual files typically requires an explicit separate action, maintaining a clear boundary between conversational exploration and committed file modifications — a boundary Agent mode intentionally loosens once you've opted into it. **Managing conversational context over time**: everything said earlier in an open Chat thread keeps shaping how later messages get interpreted, which is exactly what you want mid-debugging-session but becomes a liability the moment you pivot to something unrelated — the model can quietly carry over assumptions from the earlier topic. Cmd/Ctrl+N wipes the slate and opens a fresh thread with no memory of the previous one, and reaching for it deliberately at each real task boundary tends to keep answers sharper than letting a single sprawling conversation drift across several unconnected pieces of work. **Related shortcuts**: Cmd/Ctrl+Shift+L for adding specific code as context, @ for explicitly referencing a file or symbol, Cmd/Ctrl+I for Agent mode, Cmd/Ctrl+N for starting fresh, and Cmd/Ctrl+K for scoped inline edits better suited to smaller, well-defined changes. **Working across a monorepo**: in a large multi-package repository, Chat's automatic retrieval can sometimes pull context from an unrelated package that happens to share similar naming conventions. Explicitly scoping a question with an @ file or folder reference becomes considerably more valuable in this kind of codebase than in a small single-purpose repo, where the AI's automatic guess is far more likely to already be correct. **Saving useful conversations**: for a particularly useful debugging session or architectural discussion, it's worth manually copying the key conclusions out of Chat into a comment, commit message, or team wiki before starting a new conversation, since Cursor's chat history, while browsable, isn't a substitute for durable project documentation that teammates without access to your specific chat history will also need to read later.