Code intelligence
How the agent finds definitions, references and type errors with your language servers, falls back to parsing when none is available, and checks every edit.
Copy & share
On this page4 sections
Empryo gives the agent the same code intelligence your editor has: go to definition, find references, rename, type information and error checking. The agent looks things up by symbol instead of grepping through whole files, which is cheaper and more accurate.
There is nothing to configure. Empryo picks the best available method for each request.
What the agent can askLink to this section
| Question | Example |
|---|---|
| Where is it defined? | "where is parseConfig defined?" |
| Who uses it? | "who calls parseConfig?" |
| Rename it everywhere | "rename useChat to useConversation everywhere" |
| What type is it? | "what's the return type of fetchUser?" |
| Any errors? | "any errors in this file?" |
| What does it call? | "what does buildTools call?" |
The agent answers these with the navigate and analyze tools.
How it finds answersLink to this section
Empryo tries the most precise method first and falls back when it isn't available:
- Language server (LSP). Exact types, workspace-wide rename and diagnostics.
- Parsing the code. Outlines, imports and symbol lookups across many languages, plus deeper structure for TypeScript and JavaScript.
- Text search. A last resort that works on any file.
If no language server is running for your language, the agent still works. The fallbacks cover most questions.
Language servers without the editorLink to this section
The agent can use language servers even when the editor is closed. In the terminal UI, when you open the embedded Neovim editor (Ctrl+E), requests go through the servers Neovim already runs. When it's closed, Empryo starts the servers itself.
See which servers are attached:
/lsp statusInstall more servers from the Mason registry:
/lsp installSee LSP for more.
Checks after every editLink to this section
Empryo records the errors in a file before an edit and compares them afterwards. The agent is told:
- which errors it just introduced,
- which errors it just fixed,
- which errors appeared in other files because of the edit.
This lets the agent fix its own mistakes right away, without running a full typecheck after every change.