Add custom Neovim chat frontend for opencode.nvim with configurable keymaps #1
Conversation
Co-authored-by: verlihirsh <6280012+verlihirsh@users.noreply.github.com>
Co-authored-by: verlihirsh <6280012+verlihirsh@users.noreply.github.com>
Co-authored-by: verlihirsh <6280012+verlihirsh@users.noreply.github.com>
|
@copilot I want to setup which frontend were going to use in configuration file. Also, add default values and keymaps |
Co-authored-by: verlihirsh <6280012+verlihirsh@users.noreply.github.com>
I've added frontend selection, default keymaps, and automatic setup. Changes in commit 7f2dc0d: Configuration:
Auto-setup:
Example: vim.g.opencode_opts = {
chat = {
enabled = true, -- Enable chat UI
keymaps = {
open = "<leader>oc", -- Or {"<leader>oc", "<C-o>"}
}
}
}Updated README with full configuration documentation. |
There was a problem hiding this comment.
Pull request overview
This PR adds a custom Neovim chat UI frontend as an alternative to the terminal TUI. The implementation provides a floating window with real-time streaming, configurable keymaps, and session management, communicating with the existing opencode server via HTTP/SSE endpoints.
Key Changes:
- New chat UI module with floating window, message rendering, and streaming support
- SSE event handler for real-time message updates
- Fully configurable chat frontend (disabled by default) with customizable keymaps, provider, model, and window dimensions
- Markdown syntax highlighting via treesitter for chat messages
Reviewed changes
Copilot reviewed 8 out of 8 changed files in this pull request and generated 16 comments.
Show a summary per file
| File | Description |
|---|---|
plugin/chat.lua |
Automatic global keymap setup when chat is enabled |
lua/opencode/ui/chat_init.lua |
Entry point coordinating server connection and session initialization |
lua/opencode/ui/chat_events.lua |
SSE event handler for message deltas, session events, and errors |
lua/opencode/ui/chat.lua |
Main chat UI implementation with window management, rendering, and user interactions |
lua/opencode/config.lua |
Configuration schema for chat options including keymaps and display settings |
lua/opencode.lua |
Exposed chat() API for plugin consumers |
ftplugin/opencode_chat.lua |
Filetype plugin for markdown syntax highlighting in chat buffers |
README.md |
Documentation for chat functionality, configuration, and keybindings |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
lua/opencode/ui/chat_events.lua
Outdated
| local current_msg = state.messages[state.streaming_message_index] | ||
| if current_msg then | ||
| current_msg.text = (current_msg.text or "") .. delta | ||
| chat.render() | ||
| end |
There was a problem hiding this comment.
The function checks if state.streaming_message_index exists but doesn't verify that the index is valid for the messages array. If the messages array has been modified (e.g., cleared), this could cause an attempt to access an invalid index. Add a bounds check: state.streaming_message_index and state.streaming_message_index <= #state.messages.
| -- Only process events if chat window is still open | ||
| local state = chat.get_state() | ||
| if not state then | ||
| M.unsubscribe() |
There was a problem hiding this comment.
Calling M.unsubscribe() when the chat window is closed could cause issues if there are multiple event subscribers or if events are needed elsewhere in the plugin. The function should only unsubscribe if this chat window was the sole subscriber. Consider adding a reference counting mechanism or letting the chat cleanup handle unsubscription explicitly.
| M.unsubscribe() |
lua/opencode/ui/chat.lua
Outdated
| table.insert(M.state.messages, { | ||
| role = "user", | ||
| text = text, | ||
| timestamp = os.time(), |
There was a problem hiding this comment.
The timestamp field is added to user messages but never used anywhere in the codebase. Either remove this unused field or implement functionality that uses it (e.g., displaying message timestamps in the UI).
| timestamp = os.time(), |
lua/opencode/ui/chat.lua
Outdated
| -- Interrupt | ||
| set_keymap(keymaps.interrupt or "<C-c>", function() | ||
| M.interrupt() | ||
| end, vim.tbl_extend("force", opts, { desc = "Interrupt" })) |
There was a problem hiding this comment.
The vim.tbl_extend call appears incorrect here. The function expects (behavior, table1, table2, ...), but { desc = "Interrupt" } should be used instead of the entire vim.tbl_extend call as the third argument to vim.keymap.set. This will cause the keymap to not work properly or throw an error.
| end, vim.tbl_extend("force", opts, { desc = "Interrupt" })) | |
| end, "Interrupt") |
lua/opencode/ui/chat.lua
Outdated
| -- Add separator | ||
| if i > 1 then | ||
| table.insert(lines, "") | ||
| table.insert(lines, string.rep("─", 80)) |
There was a problem hiding this comment.
The magic number 80 is hardcoded for the separator width. This should either be dynamically calculated based on window width, or defined as a constant at the module level with a descriptive name like SEPARATOR_WIDTH. The current implementation may look broken in narrower or wider windows.
ftplugin/opencode_chat.lua
Outdated
| local ok_ts = pcall(vim.treesitter.start, vim.api.nvim_get_current_buf(), "markdown") | ||
| if not ok_ts then |
There was a problem hiding this comment.
The variable name ok_ts is inconsistent with the naming pattern established by ok in the previous line. Both variables serve the same purpose (checking if a pcall succeeded). Consider renaming to ts_ok for consistency, or better yet, reuse ok since the previous value is no longer needed.
| local ok_ts = pcall(vim.treesitter.start, vim.api.nvim_get_current_buf(), "markdown") | |
| if not ok_ts then | |
| ok = pcall(vim.treesitter.start, vim.api.nvim_get_current_buf(), "markdown") | |
| if not ok then |
| -- Find which message the cursor is on | ||
| local cursor_line = vim.api.nvim_win_get_cursor(M.state.winid)[1] | ||
| local current_line = 0 | ||
|
|
||
| for _, msg in ipairs(M.state.messages) do | ||
| -- Account for separator and header | ||
| if current_line > 0 then | ||
| current_line = current_line + 3 -- blank, separator, blank | ||
| end | ||
| current_line = current_line + 2 -- header + blank | ||
|
|
||
| local content_lines = vim.split(msg.text or "", "\n") | ||
| local msg_end = current_line + #content_lines | ||
|
|
||
| if cursor_line >= current_line and cursor_line <= msg_end then | ||
| -- Found the message, yank it | ||
| vim.fn.setreg('"', msg.text or "") | ||
| vim.notify("Message yanked to clipboard", vim.log.levels.INFO, { title = "opencode" }) | ||
| return | ||
| end | ||
|
|
||
| current_line = msg_end | ||
| end |
There was a problem hiding this comment.
The cursor line calculation for message yanking is fragile and can easily get out of sync with the rendering logic in the render() function. If the rendering logic changes (e.g., different separator lengths, additional spacing), this calculation will break. Consider storing message line ranges during rendering or using markers/extmarks to track message boundaries more reliably.
| -- Find which message the cursor is on | |
| local cursor_line = vim.api.nvim_win_get_cursor(M.state.winid)[1] | |
| local current_line = 0 | |
| for _, msg in ipairs(M.state.messages) do | |
| -- Account for separator and header | |
| if current_line > 0 then | |
| current_line = current_line + 3 -- blank, separator, blank | |
| end | |
| current_line = current_line + 2 -- header + blank | |
| local content_lines = vim.split(msg.text or "", "\n") | |
| local msg_end = current_line + #content_lines | |
| if cursor_line >= current_line and cursor_line <= msg_end then | |
| -- Found the message, yank it | |
| vim.fn.setreg('"', msg.text or "") | |
| vim.notify("Message yanked to clipboard", vim.log.levels.INFO, { title = "opencode" }) | |
| return | |
| end | |
| current_line = msg_end | |
| end | |
| local bufnr = M.state.bufnr | |
| if not bufnr or not vim.api.nvim_buf_is_valid(bufnr) then | |
| return | |
| end | |
| -- Current cursor line (1-based) | |
| local cursor_line = vim.api.nvim_win_get_cursor(M.state.winid)[1] | |
| -- Get the current buffer contents so we can derive message ranges | |
| local buf_lines = vim.api.nvim_buf_get_lines(bufnr, 0, -1, false) | |
| local total_lines = #buf_lines | |
| -- Helper to find the first occurrence of a sequence of lines (content_lines) | |
| -- in buf_lines starting from index start_idx (1-based). Returns the start | |
| -- and end line (1-based, inclusive) or nil if not found. | |
| local function find_message_range(content_lines, start_idx) | |
| if #content_lines == 0 then | |
| return nil | |
| end | |
| for i = start_idx, total_lines - #content_lines + 1 do | |
| local match = true | |
| for j = 1, #content_lines do | |
| if buf_lines[i + j - 1] ~= content_lines[j] then | |
| match = false | |
| break | |
| end | |
| end | |
| if match then | |
| return i, i + #content_lines - 1 | |
| end | |
| end | |
| return nil | |
| end | |
| local search_start = 1 | |
| -- Walk through messages in order, locating their content in the buffer | |
| for _, msg in ipairs(M.state.messages) do | |
| local text = msg.text or "" | |
| local content_lines = vim.split(text, "\n", { plain = true }) | |
| local start_line, end_line = find_message_range(content_lines, search_start) | |
| if start_line and end_line then | |
| -- Advance search start so subsequent messages don't match earlier content | |
| search_start = end_line + 1 | |
| if cursor_line >= start_line and cursor_line <= end_line then | |
| -- Found the message under cursor; yank its raw text | |
| vim.fn.setreg('"', text) | |
| vim.notify("Message yanked to clipboard", vim.log.levels.INFO, { title = "opencode" }) | |
| return | |
| end | |
| else | |
| -- If we can't find this message's content in the buffer, skip it | |
| -- and continue searching for subsequent messages. | |
| end | |
| end |
lua/opencode/ui/chat_events.lua
Outdated
| if state.streaming_message_index then | ||
| local msg = state.messages[state.streaming_message_index] | ||
| if msg then | ||
| msg.complete = true | ||
| msg.streaming = false | ||
| end | ||
| state.streaming_message_index = nil | ||
| chat.render() |
There was a problem hiding this comment.
The same bounds checking issue exists here as well. The code accesses state.messages[state.streaming_message_index] without verifying the index is valid.
lua/opencode/ui/chat_events.lua
Outdated
| if state.streaming_message_index then | ||
| local msg = state.messages[state.streaming_message_index] | ||
| if msg then | ||
| msg.text = (msg.text or "") .. "\n\n[Error: " .. error_msg .. "]" | ||
| msg.complete = true | ||
| msg.streaming = false | ||
| end | ||
| state.streaming_message_index = nil | ||
| chat.render() |
There was a problem hiding this comment.
Same bounds checking issue as in other event handlers. The code accesses state.messages[state.streaming_message_index] without verifying the index is valid.
lua/opencode/ui/chat.lua
Outdated
| local role = msg.role == "user" and "You" or "Assistant" | ||
| local header = string.format("### %s", role) | ||
| local header_line = #lines | ||
| table.insert(lines, header) | ||
| table.insert(lines, "") | ||
|
|
||
| -- Add highlight for header | ||
| table.insert(highlights, { | ||
| line = header_line, | ||
| col_start = 0, | ||
| col_end = #header, | ||
| hl_group = msg.role == "user" and "Title" or "Special", |
There was a problem hiding this comment.
The role mapping logic only handles "user" and defaults everything else to "Assistant". This could lead to confusing display if a "system" role message is shown. The code should explicitly check for all expected roles ("user", "assistant", "system") and handle them appropriately, or at least add special formatting for system messages as indicated by the event handler that creates them.
| local role = msg.role == "user" and "You" or "Assistant" | |
| local header = string.format("### %s", role) | |
| local header_line = #lines | |
| table.insert(lines, header) | |
| table.insert(lines, "") | |
| -- Add highlight for header | |
| table.insert(highlights, { | |
| line = header_line, | |
| col_start = 0, | |
| col_end = #header, | |
| hl_group = msg.role == "user" and "Title" or "Special", | |
| local role_label | |
| if msg.role == "user" then | |
| role_label = "You" | |
| elseif msg.role == "assistant" or msg.role == nil then | |
| -- Default non-user messages to Assistant unless explicitly marked otherwise | |
| role_label = "Assistant" | |
| elseif msg.role == "system" then | |
| role_label = "System" | |
| else | |
| -- Fallback for unexpected roles | |
| role_label = tostring(msg.role) | |
| end | |
| local header = string.format("### %s", role_label) | |
| local header_line = #lines | |
| table.insert(lines, header) | |
| table.insert(lines, "") | |
| -- Add highlight for header | |
| local hl_group | |
| if msg.role == "user" then | |
| hl_group = "Title" | |
| elseif msg.role == "assistant" or msg.role == nil then | |
| hl_group = "Special" | |
| elseif msg.role == "system" then | |
| -- Use a distinct style for system messages | |
| hl_group = "Comment" | |
| else | |
| hl_group = "Normal" | |
| end | |
| table.insert(highlights, { | |
| line = header_line, | |
| col_start = 0, | |
| col_end = #header, | |
| hl_group = hl_group, |
|
@copilot apply changes based on the comments in this thread |
…validation Co-authored-by: verlihirsh <6280012+verlihirsh@users.noreply.github.com>
Applied all review feedback in commit ea848ff: Bounds Checking - Added validation for Bug Fixes:
UI Improvements:
Input Validation:
All 16 review comments addressed. |
✅ Implementation Complete
Successfully implemented a custom Neovim chat frontend for opencode.nvim based on the Copilot conversation architecture.
Summary
lua/opencode/ui/chat.lua- Chat buffer manager with floating window (361 lines)lua/opencode/ui/chat_events.lua- SSE event handler for chat UI (101 lines)lua/opencode/ui/chat_init.lua- Main entry point for chat functionality (44 lines)lua/opencode.luato expose chat APIftplugin/opencode_chat.lua, 31 lines)chat.enabled) in configurationRecent Changes (commit addressing review feedback):
Bug Fixes:
Bounds Checking:
Input Validation:
UI Improvements:
Changes: 8 files, +720 lines
Quality:
✅ No deprecated APIs
✅ Stylua formatted
✅ CodeQL passed
✅ Leverages existing infrastructure
✅ Fully documented
✅ Bounds checking and input validation
✅ All review comments addressed
Original prompt
💬 We'd love your input! Share your thoughts on Copilot coding agent in our 2 minute survey.