Espx-LS

Short for espionox language server

Espx-LS utilizes the language server protocol to provide an interface for interacting with language models. This is done through a command line tool like syntax within the comments of your code. The syntax is structured as follows: <languge comment syntax> <command> <scope> option<args>

Usage

Scopes

Additional scopes can be added manually by the user, followed up in the configuration section.

A scope is a context that is associated with a character. By default, there are two scopes:

1. Global(_)

• Is initialized with just the default assistant system prompt
• Only changes when user either explicitly adds content or prompts

2. Document(^)

• Is initialized with the default assistant system prompt, and the entirety of the document it is associated with.
• Will change based on user's current document

  NOTE: Using the push command (+) with the document scope is redundant because the entirety of a current document is already included in the model's context

Commands

Currently there are two supported commands:

1. Prompt(@)

• Description: Use this command to prompt the model within the specified scope.
• Usage: @<scope> your prompt here
• Example:

  // @_ How do I read from stdin?

2. Push(+)

• Description: This command allows you to push a block of code into the model's context within the specified scope.
• Usage: +<scope> (At the top of a block of code)
• Example:

  // +_
  pub struct SomeStruct {
    id: Uuid,
    content: String,
  }
  impl SomeStruct {
    fn new() -> Self {
      Self {
        id: UUid::new_v4(),
        content: String::new(),
      }
    }
  }
  
  pub struct OtherStruct;

  NOTE: In the example above, only the SomeStruct definition and its impl block will be pushed to the model's context. This is because the Push command only includes the code block that immediately follows it. Code blocks are separated by blank lines.

Configuration

In order to get the LSP to attach within one of your projects, you must create an espx-ls.toml file in the root of the project. The [model] section is required, all other sections are optional.

[model]

• provider: either Anthropic or OpenAi
• api_key: an api for the corresponding provider Example:

[model]
provider = "Anthropic"
api_key = "your_api_key_here"

[database]

Include this if you would like to use the surrealdb integration. This will CREATE a database instance in the root of your project in the .espx-ls directory and does not require that you set one up yourself.

• namespace: the namespace of the database
• database: the name of the database
• user: username for database access
• pass: password for database access Example:

[database]
namespace = "espx"
database = "espx"
user = "root"
pass = "root"

[scopes]

for defining custom scopes

Example:

[scopes]
  [scopes.c]
  [scopes.b]
    sys_prompt = "Your prompt for scope B"

Note: In the example above, scope c will use the default assistant prompt, while scope b will utilize the specified system prompt. Both scopes can be accessed like any other scope. For instance, to prompt the model in scope c, you would use: @c your prompt.

IDE setup

As of right now I only know how to get this working in NeoVim ¯_(ツ)_/¯

I'm working on a VsCode integration, if you would like to help feel free to contact me at voidkandy@gmail.com

NeoVim Setup

First, manually compile espx-ls and put it in your PATH. Assuming you have lspconfig, the below snippet should work. You can set filetypes to any filetypes you want.

local lsp_config = require 'lspconfig'
local configs = require 'lspconfig.configs'


if not configs.espx_ls then
  configs.espx_ls = {
    default_config = {
      name = 'espx_ls',
      autostart = true,
      cmd = { 'espx-ls' },
      filetypes = { 'text', 'rust' },
      root_dir = function()
        return vim.fs.dirname(vim.fs.find({ 'espx-ls.toml' }, { upward = true })[1])
      end
    },
  }
end

lsp_config.espx_ls.setup {}

As you can see above, as of right now, espx-ls requires that you have a espx-ls.toml in your project's root in order for the LSP to know to attach.

window/showMessage

This is fully optional, but creates a slightly nicer user experience

Ensure your config has a way to handle window/showMessage requests from an LSP. NeoVim does support this out of the box, but in a way that isn't very condusive to great UX. If you have the notify plugin and this snippet of code in your config should do the trick:

vim.lsp.handlers['window/showMessage'] = function(_, result, ctx)
  local notify = require 'notify'
  notify.setup {
    background_colour = '#000000',
    render = 'wrapped-compact',
    timeoute = 100,
  }
  notify(result.message)
end

From here you should be good to go!

If you have any questions, suggestions, or anything at all feel free to reach out to me at voidkandy@gmail.com

Thanks to thePrimagen for making his HTMX LSP Open Source so I could fork it and build it into this :D