noob — agentic coding CLI

A Claude-Code-style coding agent that lives in your terminal, powered by Noob Demon (34 models across 8 providers — GPT, Claude, Gemini, DeepSeek, Grok, Qwen, Kimi, Llama).

It reads and edits files, runs shell commands, and iterates on a task — asking permission before anything destructive — all inside a polished terminal UI.

  ███╗   ██╗ ██████╗  ██████╗ ██████╗
  ████╗  ██║██╔═══██╗██╔═══██╗██╔══██╗
  ██╔██╗ ██║██║   ██║██║   ██║██████╔╝
  ██║╚██╗██║██║   ██║██║   ██║██╔══██╗
  ██║ ╚████║╚██████╔╝╚██████╔╝██████╔╝
  ╚═╝  ╚═══╝ ╚═════╝  ╚═════╝ ╚═════╝

Install

cd "noob cli"
npm install
npm link          # optional: makes `noob` available globally

Requires Node.js ≥ 18 (tested on 22).

Xác thực & gói cước (Authentication)

noob đi qua một gateway (Cloudflare Worker claude-code-proxy) để ẩn backend thật và quản lý API key qua Supabase. Bạn cần một API key để dùng:

noob login nk_xxx_xxxxxxxx     # đăng nhập, key lưu ở ~/.noob/config.json
noob usage                     # xem hạn mức còn lại
noob logout                    # đăng xuất

Trong phiên: /login <key>, /usage, /logout.

Các gói:

| Gói | Hạn mức | | --------- | ----------------------------------------- | | pro | 5 000 request / 5 giờ (cửa sổ trượt) | | proplus | 10 000 request / 5 giờ | | admin | không giới hạn | | trial | 200 request dùng thử, hết là key dead |

Mỗi lệnh gọi mô hình (kể cả từng bước tool trong một tác vụ) tính là 1 request.

Cấp/huỷ key (admin) — trong repo gateway D:\Dev\free claude code:

node scripts/admin.mjs create pro "khách A"
node scripts/admin.mjs list
node scripts/admin.mjs revoke nk_pro_xxx

Use

noob                                   # interactive session
noob "add input validation to api.js"  # start with a request
noob -m gateway-claude-opus-4-7        # pick a model
noob --yolo                            # auto-approve edits & commands

Without npm link, run node bin/noob.js ….

In-session commands

| Command | What it does | | --------------- | ------------------------------------------------------------------ | | /model [name] | switch model (fuzzy match), or list all | | /models | list every model grouped by provider | | /merge | toggle Merge AI — synthesizes GPT-5.5 + Claude + Gemini via o3 | | /search | toggle web search mode | | /chat | back to normal chat mode | | /yolo | toggle auto-approve for file/command tools (or Shift+Tab) | | /clear /new | wipe conversation context | | /status | show current model + working directory | | /exit | quit (Ctrl+C once = stop turn, twice = quit) |

How it works

The Noob Demon gateway is a stateless single-message endpoint with no native function-calling. noob layers an agent on top:

1. The full transcript (system prompt + history + tool results) is serialized into one message and streamed to the gateway's /api/chat.
2. The model replies with either a final answer or a single ```tool JSON block.
3. noob parses the tool call, asks permission if it's destructive, executes it, and feeds the result back — looping until the model answers without a tool block.

Architecture (high-level)

flowchart LR
  User[User<br/>terminal] -->|prompt| REPL[repl.js<br/>input loop + slash commands]
  REPL -->|message + history| Agent[agent.js<br/>tool-loop driver]
  Agent -->|/api/chat| API[api.js<br/>gateway client]
  API -->|HTTPS stream| GW[(Noob Demon<br/>Gateway)]
  GW -->|delta tokens| API
  API -->|onDelta callbacks| Agent
  Agent -->|parse tool block| Tools[tools.js<br/>read/write/edit/run]
  Tools -->|fs / spawn| FS[(workspace)]
  Tools -->|result text| Agent
  Agent -->|final answer| REPL
  REPL -->|persist| Sessions[(~/.noob/sessions/)]
  REPL -->|memory| Memory[(noob.md per project)]
  Agent -.optional.-> Subagent[subagent.js<br/>spawn_agent]
  Subagent -->|isolated context| API

Tools the agent can call

read_file · write_file · edit_file · list_dir · glob · grep · run_command

Destructive tools (write_file, edit_file, run_command) prompt for approval (y / n / always) unless /yolo is on.

Configuration

| Env var | Effect | | --------------------- | ---------------------------------------------------------------------------------------------------------------------------------------- | | NOOB_API_BASE | override the gateway URL | | NOOB_API_KEY | API key (overrides ~/.noob/config.json) | | NOOB_INSECURE_TLS=1 | disable TLS verification — last resort for machines behind a TLS-intercepting proxy. Prefer adding your proxy CA to the trust store. |

Model compatibility

The agent drives tools through a text protocol (the proxy has no native function-calling). Models differ in how willingly they follow it:

| Provider | Agentic tools | Notes | | ---------------------- | ---------------- | ---------------------------------------- | | Anthropic (Claude) | ✅ best | default — gateway-claude-opus-4-7 | | DeepSeek | ✅ works | good free alternative | | OpenAI (GPT/o-series) | ⚠️ often refuses | replies "I can't access your filesystem" | | Google (Gemini) | ⚠️ often refuses | same |

Stick with Claude or DeepSeek for file edits & commands. Any model is fine for plain chat, /merge, and /search.

Project structure

src/
├── api.js              # gateway client + stream parser + memory token
├── agent.js            # tool-loop driver, summarization, prompt assembly
├── tools.js            # read/write/edit/list_dir/glob/grep/run_command + bg
├── repl.js             # input loop, slash commands, session state (88 KB)
├── repl/
│   ├── complete.js     # SLASH catalog + autocomplete
│   ├── todos.js        # parseTodosFromHistory (pure)
│   ├── ultra.js        # ULTRA mode constants + prompt templates
│   └── workflow-commands.js  # /workflow help/list/load/delete (pure)
├── subagent.js         # spawn_agent / spawn_agents — sub-agent isolation
├── tui.js              # terminal UI (input, status bar, prompt rendering)
├── ui.js               # color helpers, markdown renderer, banner
├── config.js           # ~/.noob/config.json (api key, gateway, model)
├── sessions.js         # ~/.noob/sessions/ — save/list/resume
├── skills.js           # ~/.noob/skills/ — user-defined skill loader
├── workflows.js        # ~/.noob/workflows/ — user-saved workflows
├── workflows-builtin.js # ship-with-binary workflows (deep-research, etc)
├── memory.js           # noob.md per-project agent memory
├── tokens.js           # local token counter (o200k_base / cl100k_base)
├── models.js           # 34-model catalog + fuzzy resolver
├── i18n.js             # Vietnamese string table
├── update.js           # `noob update` self-updater
└── prompts/
    └── system.md       # base system prompt template

tests/                   # vitest unit tests (96 tests)
scripts/                 # check-imports, notify-discord, release.ps1

Development

npm test                 # run vitest unit tests (96 tests across 7 files)
npm run lint             # eslint --check
npm run lint:fix         # eslint --fix
npm run format           # prettier --write
npm run format:check     # prettier --check
npm run check            # lint + format:check + test (pre-commit gate)

Husky chạy lint-staged + npm test trước mỗi commit. Bỏ qua bằng git commit --no-verify nếu thực sự cần (vd commit WIP).

Troubleshooting

"CLI tự thoát sau khi hỏi quyền" — thường là readline trên Windows phát sự kiện close từ tiến trình con chạm vào console. Đã có workaround trong repl.js (input layer tự re-arm). Nếu còn xảy ra, mở issue kèm OS + Node version.

"TLS error / certificate" — máy bạn sau proxy chặn TLS (Zscaler, Cisco Umbrella…). Ưu tiên add CA của proxy vào trust store. Cuối cùng mới dùng NOOB_INSECURE_TLS=1 (tắt verify TLS toàn process, MITM-vulnerable).

"Model OpenAI/Gemini từ chối tool" — bình thường. GPT-o-series và Gemini thường refuse access filesystem qua text protocol. Dùng /model claude-opus hoặc /model deepseek-v3 cho tasks cần tool.

"Token count có vẻ sai" — tokens.js chọn encoder theo model id (o200k_base cho GPT-4o/5/o-series, cl100k_base cho phần còn lại). Đếm chính xác cho OpenAI, xấp xỉ ±5-15% cho Claude/Gemini/khác (các provider này không publish tokenizer public).

Notes & limits

• Free shared proxy: rate limits and occasional hiccups are expected.
• Context is sent in full each turn, so very long sessions cost more tokens — use /clear to reset.
• This is an unofficial tool built against a public community endpoint; it is not affiliated with Anthropic or the model providers.

License

MIT