Neovim integration for ByteAsk — the Codex-style agentic coding harness specialized for C/C++ (compiler, disassembler, and debugger as first-class agent tools).

This plugin drives the byteask CLI from inside Neovim:

• Interactive TUI in a floating window, side panel, or tab.
• Headless commands — exec, review, apply — that stream results into a scratch buffer and can apply the produced diff straight to your working tree.
• Context sends — pipe the current selection, file, or LSP diagnostics into a one-shot exec (e.g. "fix these compiler errors").
• Session control — resume/fork previous sessions.
• :checkhealth byteask and a lualine statusline component.

Inspired by codex.nvim, rebuilt for ByteAsk's richer CLI surface (exec --json, review, apply, resume/fork).

• Neovim 0.9+
• The byteask CLI on your $PATH:

  pip install --upgrade byteask   # or: pipx install byteask
  byteask login                   # authenticate once
  byteask doctor                  # verify install/auth/runtime

{
  'ByteAsk/byteask.nvim',
  cmd = { 'ByteAsk', 'ByteAskExec', 'ByteAskReview', 'ByteAskResume' },
  keys = {
    { '<leader>bb', function() require('byteask').toggle() end, mode = { 'n', 't' }, desc = 'Toggle ByteAsk' },
    { '<leader>bx', ':ByteAskExec ', mode = 'n', desc = 'ByteAsk exec' },
    { '<leader>bx', ":'<,'>ByteAskExec ", mode = 'v', desc = 'ByteAsk exec (selection)' },
    { '<leader>bd', function() require('byteask').fix_diagnostics() end, desc = 'ByteAsk: fix diagnostics' },
    { '<leader>br', function() require('byteask').review() end, desc = 'ByteAsk review' },
  },
  opts = {
    layout = 'float',       -- 'float' | 'panel' | 'tab'
    -- model = 'gpt-5.5',   -- omit to use the ByteAsk default
    -- auto_apply = true,   -- apply exec diffs to the working tree automatically
  },
}

use {
  'ByteAsk/byteask.nvim',
  config = function() require('byteask').setup{} end,
}

Inside the ByteAsk window, <C-q> hides it (the session keeps running in the background — reopen with :ByteAsk).

Defaults (pass any subset to setup/opts):

require('byteask').setup({
  cmd = 'byteask',            -- string or argv list; may include flags e.g. 'byteask --oss'
  args = {},                  -- extra CLI args always appended
  model = nil,                -- -m MODEL (nil = ByteAsk default)
  config = {},                -- -c key=value TOML overrides, e.g. { model_reasoning_effort = 'high' }

  layout = 'float',           -- 'float' | 'panel' | 'tab'
  width = 0.85, height = 0.85,-- float size (fraction of editor)
  panel_side = 'right',       -- 'left' | 'right'  (layout='panel')
  panel_width = 0.42,
  border = 'rounded',         -- 'single'|'double'|'rounded'|'none'|custom table
  start_insert = true,        -- enter terminal insert mode on open

  keymaps = { toggle = nil, quit = '<C-q>' },

  autoinstall_hint = true,    -- guide install when `byteask` is missing
  install_cmd = { 'pip', 'install', '--upgrade', 'byteask' },

  auto_apply = false,         -- run `byteask apply` after a successful exec
  notify_on_exit = true,      -- vim.notify when headless runs finish
})

require('lualine').setup({
  sections = { lualine_x = { require('byteask').status() } },
})

• The interactive TUI runs via termopen({ 'byteask', ... }) in a managed window.
• Headless commands shell out to byteask exec / review / apply with jobstart, streaming stdout/stderr into the byteask://output scratch buffer.
• resume/fork re-launch the TUI with the corresponding subcommand (model/config overrides are omitted so the restored session keeps its own recorded settings).

• Structured integration via byteask app-server (--remote) instead of terminal scraping — inline diffs, approvals, and streaming tool calls rendered natively.
• Consume byteask exec --json (JSONL events) for a real progress/tool-call UI.

See AGENTS.md — the release process (tag → verify), required secrets and where to get them, and a CI gotcha worth reading before touching .github/workflows/.

Apache-2.0