delphi.nvim

A tasteful LLM plugin.

Features

• Clean and snappy Vim buffer based chat interface
• Local storage for chat history, with Telescope integration
• Code rewrite command with unified diff
• Live streaming of tokens
• Pure Lua OpenAI client and diff algorithm

Setup

You'll need a plugin manager (such as lazy), and an OpenAI compatible LLM API (such as OpenRouter).

Example configuration with lazy.nvim:

{
        "nathom/delphi.nvim",
        keys = {
                { "<leader><cr>", "<Plug>(DelphiChatSend)", desc = "Delphi: send chat" },
                { "<leader>r", "<Plug>(DelphiRewriteSelection)", mode = "x", desc = "Delphi: rewrite selection" },
                { "<leader>a", "<Plug>(DelphiRewriteAccept)", desc = "Delphi: accept rewrite" },
                { "<leader>R", "<Plug>(DelphiRewriteReject)", desc = "Delphi: reject rewrite" },
                { "<Esc><Esc>", "<Plug>(DelphiPromptCancel)", mode = { "n", "i" }, desc = "Delphi: cancel prompt" },
        },
        opts = {
                chat = { default_model = "gemini_25" },
                rewrite = { default_model = "kimi_k2" },
                models = {
			gemini_25 = {
				base_url = "https://openrouter.ai/api/v1",
				api_key_env_var = "OPENROUTER_API_KEY",
				model_name = "google/gemini-2.5-pro",
			},
			kimi_k2 = {
				base_url = "https://openrouter.ai/api/v1",
				api_key_env_var = "OPENROUTER_API_KEY",
				model_name = "moonshotai/kimi-k2",
			},
		},
	},
	dependencies = { "nvim-lua/plenary.nvim" },
}

In this case, delphi will expect your OpenRouter API keys under the environment variable OPENROUTER_API_KEY.

Usage

Chat

Type :Chat to open up a chat buffer. If one is already open in your Neovim instance, it will switch to that buffer. If you want to create a new chat regardless, use :Chat new.

Example blank chat buffer:

---
model: kimi_k2
temperature: 0.7
---

System:


User:

To send a message to the model, type text below User:, switch to normal mode, and trigger <Plug>(DelphiChatSend).

---
model: kimi_k2
temperature: 0.7
---

System:


User:
Is it worth switching to Neovim?

You should see the model's response being streamed in live.

---
model: kimi_k2
temperature: 0.7
---

System:


User:
Is it worth switching to Neovim?

Assistant:
Short answer  
If you already know Vim and are happy with it, you probably don’t *need* to switch—but if you enjoy experimenting,
want modern features without waiting for Bram to merge them, or are starting from scratch, Neovim is almost always the better choice today.

Below is a practical “decision matrix” so you can decide whether the switch is worth the one-time migration cost for *you*.
...

The system prompt can usually be left empty, as the provider will set a reasonable one for you.

Chat history

You have two options:

• Use :Telescope delphi chats
• Run :Chat list to view the ids and titles of chats, and :Chat go <id> to open them

Rewrite

This command allows the LLM to rewrite a selected block of code according to your instructions. It will display a unified diff, which is updated as tokens are streamed in.

To use it

• open a buffer with some text
• highlight a few lines in Visual Lines mode (shift-V)
• trigger <Plug>(DelphiRewriteSelection), which should open a popup
• instruct the model, hit ENTER
• accept or reject the changes via <Plug>(DelphiRewriteAccept) or <Plug>(DelphiRewriteReject)

Configuration

These are the schema:

---@class Model
---@field base_url string
---@field api_key_env_var string
---@field model_name string
---@field temperature number

---@class Config
---@field models table<string, Model>
---@field allow_env_var_config boolean
---@field chat { system_prompt: string, default_model: string?, headers: { system: string, user: string, assistant: string } }
---@field rewrite { system_prompt: string, default_model: string?, prompt_template: string }

These are the default opts:

opts = {
	models = {},
	allow_env_var_config = false,
        chat = {
                system_prompt = "",
                default_model = nil,
                headers = {
                        system = "System:",
                        user = "User:",
                        assistant = "Assistant:",
                },
        },
        rewrite = {
                default_model = nil,
                system_prompt = [[
You are an expert refactoring assistant. You ALWAYS respond with the rewritten code or text enclosed in <delphi:refactored_code> tags:
<delphi:refactored_code>
...
</delphi:refactored_code>]],
                prompt_template = [[
Full file for context:
<delphi:current_file>
{{file_text}}
</delphi:current_file>

Selected lines ({{selection_start_lnum}}:{{selection_end_lnum}}):
<delphi:selected_lines>
{{selected_text}}
</delphi:selected_lines>

Instruction: {{user_instructions}}. Return ONLY the refactored code inside <delphi:refactored_code> tags. Preserve formatting unless told otherwise. Try to keep the diff minimal while following the instructions exactly.]],
        },
}
```