✨ Why WenOX CLI

• Actually agentic. One prompt isn't one answer. The assistant keeps calling tools — read, search, edit, run — until the task is finished, then reports back.
• Built for the terminal. Full-screen interface with a framed prompt, a chat transcript you can scroll, a status bar and a command palette. No browser tab.
• You stay in control. Command execution asks first. Anything outside your project asks first. Plan mode changes nothing at all.
• A real language server. Go-to-definition, references and hover come from an LSP server, not from grep.
• Remembers where you were. Every conversation is saved and given a title; pick it up later exactly where you left off.
• Speaks your language. Turkish and English interface, with the model replying in whichever language you write in.
• Extensible. MCP servers (stdio or HTTP) plug in as tools, and you can paste an image from the clipboard straight into the conversation.
• Stays cheap in long sessions. Requests are built so that the provider's prefix cache hits; the part that doesn't change is never recomputed.

📦 Install

npm install -g @wenox/cli

Node.js 20 or newer. Everything ships with the package — no extra steps, no native build.

🚀 First run

Start wenox and it walks you through setup:

1. Language — Turkish or English (Enter keeps the detected one)
2. API key — paste the key; Ctrl+O opens the key page in your browser
3. Verification — the key is checked before it is ever saved
4. Welcome — your name, subscription and remaining credits

Get a key at me.wenox.co/api-key.

🖥️ Usage

wenox                                   # interactive session
wenox -p "how many files are in here?"  # one-shot answer, then exit
wenox -m <id> -y                        # pick a model, auto-approve commands
wenox -s ses_f78edf902ffe               # resume a saved session
wenox -c                                # continue the most recent session
wenox stats                             # token usage summary

| Option | Description | | --- | --- | | -m, --model <id\|no> | Model to use — see /model for the list | | -k, --key <key> | WenOX API key (saved permanently) | | -s, --session <id> | Resume a saved session | | -d, --cwd <path> | Starting working directory | | -y, --auto-approve | Run commands without asking | | -p, --prompt <text> | Run a single prompt and exit | | -c, --continue | Continue the most recent session | | --fork | Fork the most recent session into a new one | | --format json | With -p, print the result as JSON | | --debug | Write a diagnostic log (~/.wenox/debug.log) | | -v, --version | Show version | | -h, --help | Show help |

There are also commands that never open the TUI:

| Command | Description | | --- | --- | | wenox models [--format json] | List models (context window, vision support) | | wenox sessions | List past sessions | | wenox session delete <id> | Delete a session | | wenox stats [--format json] | Token usage summary | | wenox mcp list \| add \| remove | Manage MCP servers |

⌨️ Interface

A command renders as a card — the $ line, then its output. Long output is shortened to six lines; click the card to see the rest, click again to collapse it.

The prompt never locks. While the assistant is replying you can keep typing — your message is queued and sent the moment the current turn ends.

| Key | Action | | --- | --- | | Enter | Send | | / | Command menu — filters as you type, ↑/↓ to pick | | @ | File mention menu — inserts the file into the message as a chip | | !command | Runs the command directly, without going to the model | | Tab | Switch mode: Build / Plan | | Ctrl+P | Command palette | | Ctrl+V / Alt+V | Attach the image from your clipboard | | Esc | Close a menu · cancel the reply or the running command | | Ctrl+C | Works anywhere: closes a panel, cancels, or exits | | ↑ / ↓ | Input history (your draft is restored) | | Ctrl+A / Ctrl+E | Start / end of line | | Ctrl+U / Ctrl+K | Delete to line start / line end | | Ctrl+W | Delete the previous word | | Alt/Option + ← → | Jump by word | | PgUp / PgDn | Scroll the conversation | | Mouse wheel | Scroll the conversation |

Drag with the mouse to select text, then Ctrl+C to copy it. Prefer your terminal's own selection and scrolling? Start with WENOX_NO_MOUSE=1.

On wide terminals a panel sits on the right: context usage, open tasks, plan limits and the files changed in this session. It is not drawn on narrow terminals; tune it from the config (sidebar, sidebarMinColumns).

🧰 Capabilities

The assistant works on your project through these tools:

read_file · write_file · edit_file · list_dir · search_code · glob · run_command · code_intel · webfetch · ask_user · todo_write

code_intel uses a real language server:

| Language | Server | Install | | --- | --- | --- | | TypeScript / JavaScript | typescript-language-server | npm i -g typescript-language-server | | Python | pyright-langserver | npm i -g pyright | | Go | gopls | go install golang.org/x/tools/gopls@latest | | Rust | rust-analyzer | rustup component add rust-analyzer |

If a server isn't installed the tool says so and tells you how to get it — it never just fails.

ask_user lets the assistant ask a short multiple-choice question when a real decision is needed. It won't nag you with questions it can answer itself.

Paste images. Ctrl+V (or Alt+V on Windows Terminal) attaches the image on your clipboard to the message: screenshots, design frames and error pictures are read directly. /model shows which models support images; with one that doesn't, the CLI warns you.

Large files are written in pieces. The assistant doesn't try to fit a whole file into one giant output: it starts with write_file and appends section by section with edit_file. If the response hits the output limit it continues from where it stopped, and a half-finished tool call is never executed.

Your project rules are read. An AGENTS.md (or CLAUDE.md) in the project root and in parent directories is added to the system prompt. If you don't have one, /init inspects the project and writes it for you.

MCP servers. Connect one over stdio with wenox mcp add <name> <command> [args], or over HTTP by setting url in the config; its tools are added to the assistant's tool list.

🔒 Safety

• Commands ask first. Execution is confirmed before it happens; -y opts out.
• Outside your project asks first. Choose Allow once, Allow always (remembered per project) or Reject.
• Unsafe folders are flagged. Start in a home directory, a drive root or a system folder and the CLI warns you, tells the assistant to be careful, and requires permission for every path — list_dir included.
• Self-destructive commands are refused. Anything that would kill every Node process (taskkill /IM node.exe, pkill node, …) is blocked, because the CLI itself is one of them.
• Esc stops the work. A running command is killed together with its child processes — a forgotten dev server included.
• You can write permission rules. Put entries like { "tool": "run_command", "pattern": "git *", "action": "ask" } into the permissions list in the config: deny never runs it, allow never asks, ask asks every time.
• Secret files ask separately. Reading or writing a file like .env needs permission even when it is inside your project.

🗺️ Modes

The active mode sits in the status bar; Tab switches.

| Mode | Behaviour | | --- | --- | | Plan — default | Read-only. The writing and command tools are not in the tool list at all, so the assistant never even attempts them. It inspects and proposes a plan. | | Build | Reads, edits files and runs commands to get the job done. |

Start in Plan, see what the assistant intends to do, then Tab into Build to let it happen. The current mode is also repeated on every request, so older messages in the conversation can't mislead it.

🌍 Languages

Turkish and English, detected from your system and switchable with /lang or WENOX_LANG.

Only the interface is translated. Prompts and tool schemas stay in English, and the assistant answers in whichever language you write in.

💾 Sessions

Conversations are saved automatically and titled after a few exchanges. On exit the terminal shows how to come back:

  Session   Greeting
  Continue  wenox -s ses_f78edf902ffe

Resuming restores the messages, the token counter, the model, the task list and the working directory. /sessions lists everything and loads whichever you pick; wenox -c continues the most recent session and --fork copies it into a new one. From the terminal you can also list them with wenox sessions and drop one with wenox session delete <id>.

⚙️ Commands

Everything is driven from inside the session — change the model, the language or the key, inspect the account, manage the context. Settings can also be supplied through environment variables: WENOX_API_KEY, WENOX_DEFAULT_MODEL, WENOX_LANG, WENOX_API_BASE_URL, WENOX_REQUEST_TIMEOUT_MS, WENOX_DEBUG.

| Command | Description | | --- | --- | | /help | Commands and shortcuts | | /model | Change model (listed with context window and vision support) | | /lang | Change interface language | | /key | Update the API key (verified before saving) | | /me | Account, plan and limits (5-hour / weekly / monthly) | | /theme | Pick a theme | | /keys | Show the shortcuts | | /editor | Compose the message in your $EDITOR | | /init | Inspect the project and write AGENTS.md | | /compact | Summarise the context to free space | | /new | Clear the context | | /sessions | List and load past sessions | | /auto | Toggle auto-approve | | /status | Session status | | /exit | Quit |

Plan limits (5-hour / weekly / monthly) show up in the status bar and the side panel with their percentage and reset time.

🧩 OpenCode integration

You can use WenOX inside OpenCode too — no need to touch OpenCode's provider screen:

wenox opencode-setup      # asks for the key, fetches models, writes the config
wenox opencode-sync       # refreshes the model list
wenox opencode-remove     # removes the WenOX provider from OpenCode
wenox opencode-api-key    # updates the key

• Only the provider.wenox-go section is added to ~/.config/opencode/opencode.json; your other providers, MCP servers and permissions are preserved. If the file is not valid JSON it is left untouched.
• The model list comes from /v1/models: the name, context/output limits, image (attachment), reasoning and tool-calling (tool_call) support are written per model.
• The API key is never written into opencode.json: it goes to ~/.config/wenox/api-key and the config only holds a {file:...} reference, so sharing your config does not leak the key.
• The same commands work on Windows, macOS and Linux.

Restart OpenCode afterwards.

🛠️ Development

Written in TypeScript and compiled with tsc. The published package ships the compiled output, so installing it needs nothing extra.

npm install
npm run build   # compile
npm start       # compile, then run the CLI

📄 License

MIT