natureco-cli

NatureCo CLI

The power of AI, now at your fingertips. Discover the speed of the terminal with NatureCo.

Most AI coding CLIs lock you into one vendor's model and stop at your editor. NatureCo doesn't. One agent, 12 providers to choose from, that writes and ships your code, remembers what matters across sessions, and — uniquely — runs your Telegram, Discord, Slack, WhatsApp, and iMessage bots from the same install. Swap providers without losing memory. Automate without duct tape. Never send another risky rm -rf without a confirmation prompt.

A Claude Code & OpenClaw alternative, built to not box you in: Multi-agent orchestration · Cross-session memory backed by a real git-native memory engine (Urðr) · Token-budgeted context · Dangerous-command approval on every shell path · 12 providers, 133 models · 91 tools · 103 commands · 14 channels (9 stable messaging + 4 experimental + webhooks).

Why teams pick it over the alternatives:

• 🔓 No vendor lock-in — switch between OpenAI, Anthropic, Gemini, MiniMax, or 8 more providers with one command; your memory and workflows carry over.
• 🧠 Memory that's actually durable — backed by a git-native, hash-chained architecture, not a JSON blob that silently corrupts.
• 💬 It lives where your team already talks — the same agent answers in your terminal, your Telegram bot, and your Slack workspace.
• 🛡️ Guardrails by default — every shell command is risk-scored before it runs; nothing destructive executes without your say-so.
• 🌍 Genuinely cross-platform — audited and live-tested on macOS, Windows, and Linux, not just "should work."

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

---

🚀 Quick Start

# 1. Install
npm install -g natureco-cli

# 2. First-run setup wizard (provider, model, bot name)
natureco setup

# 3. Start chatting
natureco chat

# 4. Or launch the coding agent
natureco code

Ready in 30 seconds. On first launch, the wizard walks you through it: pick a provider → enter your API key → choose a model → name your bot. That's it.

Cross-platform: works natively on macOS, Windows, and Linux. Core tools (file, search, shell, git, code execution, web, memory) are OS-agnostic; platform-specific helpers degrade gracefully.

---

🆕 What's New

| Version | Highlights | |---------|-----------| | v5.71.0 | ~65% fewer tokens per request, one coding agent, and safety gates that actually fire. Tool schemas now load on demand (13.8k → ~4.5k tokens per request; a name catalogue plus enabletools keeps every tool reachable), and the per-message workflow planner is opt-in — the agent loop drives directly. code absorbs the legacy agent (/run /test /git /commit /index /memory), gains MCP servers, resumable sessions (-c / --resume) and headless -p. Several gates turned out never to run: chat executed rm -rf with no prompt, --dry-run was bypassable via codeexecution, --dir didn't change the working directory, and context compaction was a no-op — all fixed and shared across every surface. Adds Matrix, Teams, Google Chat and Zalo (14 channels, these four experimental/unverified). 995 → 1125 tests. | | v5.70.0 | natureco code visual + experience overhaul: answers now stream live with incremental markdown rendering and syntax-highlighted code, Esc interrupts an in-flight turn transactionally (session stays intact), a chat-style bordered input box replaces the bare prompt, tool calls print unified cards with real colored diffs and secret redaction, plus a live model · tokens · elapsed status line — with a critical 256-color fix so the palette finally renders on every terminal. 936 → 995 tests. | | v5.69.0 – v5.69.4 | Windows/macOS GUI automation reliability + real CI enforcement: fixed AppleScript injection and locale-ambiguous date parsing across 6 macOS automation tools, a Windows SendKeys assembly-loading bug, and a Windows timeout command that failed under piped stdin — all found via genuine live-device testing, not just automated tests. | | v5.68.0 – v5.68.6 | Full command-surface security audit: every src/commands/ file (107 files) audited and fixed — credential masking, path-traversal validation, honest success/failure reporting (commands no longer silently claim success on failure), plus full test coverage added for all 91 built-in tools. | | v5.66.0 – v5.67.6 | Real Urðr memory engine integration: natureco's persistent memory (memory write, memory search, code session context) is now genuinely backed by Urðr's git-native, hash-chained memory architecture — not just an in-repo JSON file — with automatic legacy fallback. Also: 10 shell-injection fixes converting execSync string-building to safe execFileSync argv arrays, and a Turkish-locale filename bug fix across 25 files. | | v5.65.6 | Survives the just-launched-app race: GUI actions right after opening an app no longer fail on a transient kAXErrorFailure (-25200) — osaScript retries once, and macappopen waits for the UI to settle. | | v5.65.5 | Fewer false-negative timeouts: osascript timeout raised from 10s to 20s so slow-but-successful GUI actions (typing, saving) aren't misreported as failures. | | v5.65.4 | Accessible mouseposition errors: the raw "mouse variable undefined" AppleScript error is now classified as a missing Accessibility permission, with the same actionable guidance as every other GUI action. | | v5.65.3 | Real Windows mouse/scroll automation: click and scroll now use a real user32.dll mouse event instead of a non-existent SendKeys code, and macOS screenshot permission failures are always detected. | | v5.65.2 | Reliable macOS automation recovery: simple open requests stop after success, stale browser profiles retry once, and Screen Recording/Accessibility failures provide precise Cupertino Terminal permission guidance. | | v5.65.0 | Persistent browser agent: visible system Chrome/Chromium, persistent login/storage, and reliable snapshot → @ref click/fill → verify automation replace one-shot headless browsing. |

See CHANGELOG.md for the full release history.

Slash-Prefix Command System

On iMessage and WhatsApp, only messages starting with / are processed as commands: