Neovim plugin to chat with Codex, apply AI edits, run prompts, and view exec logs.

Early scaffold: streaming chat works, prompts load, exec log streams JSON, apply replaces buffer in a diff pane. Diff hunks/markdown polish/persistence are future work.

With lazy.nvim:

{
  "yourname/codex.nvim",
  config = function()
    require("codex").setup({
      -- optional overrides
      -- model = "gpt-5.1-codex",
      -- cli = { enabled = true, command = "codex", args = { "--json" } },
    })
  end,
}

Requirements: Neovim 0.9+, plenary.curl, Codex auth (~/.codex/auth.json or env key), optional codex CLI for exec log.

• :CodexChat [prompt] open chat (prompts via vim.ui.input if omitted).
• :CodexAsk|Explain|Fix [prompt] use selection/file context; streams response.
• :CodexApply [prompt] shows unified diff in vsplit; a apply buffer, A apply all hunks, h apply current hunk, n/p navigate hunks, q close.
• :CodexPrompt <name> [ARGS] run prompt from ~/.codex/prompts/*.md ($ARGUMENTS expands).
• :CodexLog open exec event log; c/f/w toggle commands/files/web; q close.
• :CodexExec <prompt> runs codex exec --json (requires cli.enabled=true).
• :CodexExecResume resume last exec session.
• :CodexHealth run health checks.

• Chat streams via SSE; selection is appended to prompt for ask/explain/fix.
• Transcript kept in-memory per session; not persisted.
• Apply flow replaces the entire buffer; no hunk-level apply yet.
• Exec log shows agent/commands/file changes/web search with filters.

See lua/codex/config.lua:

• model = "gpt-5.1-codex", model_provider = "openai", reasoning_effort = "medium".
• base_url = "https://api.openai.com/v1", sandbox_mode = "read-only", approval_policy = "on-request".
• prompt_dir = "~/.codex/prompts", cli = { enabled = false, command = "codex", args = { "--json" } }.
• UI borders/sizes and log filters configurable.

• Markdown/scroll polish for chat/log.
• Better error surfacing and transcript persistence.

MIT