byteask.nvim

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

💬 Join the community

discord.gg/vx5Eu4YNzG — community support, direct access to the team, and the fastest way to report an issue. Direct email: anirudha@byteask.ai.

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").
• Scoped review — uncommitted changes, a base branch, a specific commit, or the whole repo, matching what byteask review actually supports.
• AGENTS.md steering — jump straight to (or scaffold) the file ByteAsk reads for repo-local instructions.
• Full session lifecycle — resume/fork, plus archive/unarchive/delete.
• Per-invocation model/config overrides without editing your Lua config.
• :checkhealth byteask and a lualine statusline component.

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

Requirements

• 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

Install

lazy.nvim

{
  '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 (scoped)' },
    { '<leader>ba', function() require('byteask').agents() end, desc = 'ByteAsk: open/create AGENTS.md' },
    { '<leader>bs', function() require('byteask').sessions() end, desc = 'ByteAsk: sessions (archive/unarchive/delete)' },
    { '<leader>bm', function() require('byteask').model() end, desc = 'ByteAsk: set model/config override' },
  },
  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
  },
}

packer.nvim

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

Commands

Command	Description
:ByteAsk [prompt]	Toggle the TUI; with args, open seeded with that prompt
:ByteAskToggle	Toggle the ByteAsk window (session survives when hidden)
:ByteAskExec {instr}	Headless exec. In visual mode, sends the selection as context
:ByteAskFixDiagnostics	Send current-buffer LSP/compiler diagnostics to exec for a fix
:ByteAskReview	Run byteask review, scoped via a prompt: uncommitted / base branch / commit / whole repo
:ByteAskApply	Apply the latest agent diff to the working tree (byteask apply)
:ByteAskResume[!]	Resume a session (picker; ! = most recent)
:ByteAskFork[!]	Fork a session (picker; ! = most recent)
:ByteAskAgents	Open the nearest AGENTS.md (walks up from the current file), or offer to create one
:ByteAskSessions	Browse all sessions, or archive/unarchive/delete one by id or name
:ByteAskModel[!]	Set a model / -c override for future invocations (! clears it)

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

Configuration

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
})

Statusline (lualine)

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

How it works

• 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).
• :ByteAskAgents never spawns byteask — it just opens/creates the file on disk; ByteAsk picks it up itself on the next turn.
• :ByteAskSessions's "browse all" delegates to byteask resume --all's own picker; archive/unarchive/delete shell the matching subcommand directly.
• :ByteAskModel stores a transient override (session-lifetime, not persisted to disk) that apply_common_flags layers on top of your static setup() config for every subsequent invocation, until cleared with :ByteAskModel!.

Roadmap

• 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.

For coding agents

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/.

License

Apache-2.0