Netro CLI — Documentation

A terminal-native AI coding agent: real-time HUD, multi-phase workflows, self-verifying loops, an extensible skills system, and a privacy layer. One standalone binary — no npm, no Node required.

Netro is proprietary software (not open-source), built and maintained by Netro Indonesia. It's safe to run: telemetry is off by default, credentials are stored hashed, and the privacy layer keeps your traffic yours. Questions or access: t.me/netro_1933.

## Install

One line. The installer downloads a standalone binary and puts netro on your PATH — no npm or Node needed.

macOS / Linux

curl -fsSL https://www.netrocli.bond/install.sh | bash

Windows (PowerShell)

irm https://www.netrocli.bond/install.ps1 | iex
The installer also fetches netro-extras (extensions, agents, skills, prompts) into ~/.netro/agent/, so gated commands like /cli-login and the web tools are available immediately.

Verify:

netro --version

## Quick start

cd your-project
netro                     # start the interactive agent

> explain what this repo does
> add a health-check endpoint and a test for it
> /help                   # list all commands

Netro reads your project files, runs tools (read/write/edit/bash/grep), and keeps a live HUD at the bottom showing model, context usage, cost, running tools, and todos.

## Subscription & login

The CLI is free with your own API keys. A subscription gives you hosted model access (no keys of your own) via an access key.

  • Subscribe on Telegram: t.me/netro_1933
  • You receive an access key like netro-xxxxxxxx…
/cli-loginPrompt for your access key and sign in
/cli-whoamiShow the currently signed-in key
/cli-logoutSign out and re-lock the CLI
/cli-revalidateForce re-check key against the registry now
Keys are verified against https://www.netrocli.bond/auth.json (hashes only). Local session is ~/.netro/agent/cli-auth.json. Every time you open netro, the CLI re-fetches the registry; if the key is missing, active: false, or past expires_at, you are auto-logged out. Binary updates do not clear login — lifetime access lasts while the operator keeps your hash active.

## Commands

Type /help in-session for the full list. Highlights:

Core

/helpList all commands
/model, /providerSwitch model / provider mid-session
/effort, /thinkingSet reasoning effort (off → max)
/hudToggle / configure the status HUD
/verifyRun verification on the last change
/brainShow the working-memory scratchpad

Workflow & session

/coordinatorOrchestrate subagents in phases
/goalSet a standing objective across turns
/checkpointSave a resumable checkpoint
/handoff, /summarizeSummarize & continue long sessions
/cronSchedule recurring runs
/export-mdExport the session to Markdown

Extend & ops

/skill-installInstall skills from a source
/agent-newScaffold a custom subagent
/learn, /curatorSave / manage learned skills
/stealthShow privacy / proxy status
/memoryDurable facts across sessions
/guardianPentest playbooks & scanner install
/updateCheck / install latest binary
/mcpMCP servers status / reload
/obedience, /freeCooperation posture / unrestricted package

## Effort / reasoning

Netro exposes seven effort levels: off · minimal · low · medium · high · xhigh · max. Set them live:

/effort high
/effort max
/effort cycle       # step to the next level

Effort is sent to the provider as a real reasoning hint. When the HUD badge shows high/xhigh/max, a pulsing marker glows to signal deep reasoning is engaged.

Force effort on any model — even ones that don't advertise thinking support — with NETRO_FORCE_EFFORT=1 (on by default). The provider decides what to do with the hint.

## Coordinator

For non-trivial tasks, netro can act as a coordinator that directs specialist subagents through phases: Research → Synthesis → Implementation → Verification.

/coordinator on     # enable orchestration discipline
/coordinator status
  • Read-only research runs in parallel.
  • Write-heavy work is serialized per file-set — subagents never clobber each other.
  • Verification runs with fresh eyes to prove changes work, not just that they exist.

Powered by the subagent tool (single / parallel / chain modes) and 50+ specialist agents.

## Web search & fetch

Two tools bring the live web into a session:

web_fetchFetch a URL and read it back as clean text
web_searchSearch the web — titles, links, snippets

Keyless search works out of the box (via Bing RSS). For higher-volume, dependable results, set an API key:

# any one of these enables premium search
export BRAVE_API_KEY=...      # brave.com/search/api (free tier)
export SERPER_API_KEY=...     # serper.dev
export TAVILY_API_KEY=...     # tavily.com

## Guardian / pentest

Netro can orchestrate Guardian-style security assessments without shipping 50 scanner binaries in the install zip. Same external CLIs Guardian wraps (nmap, httpx, nuclei, subfinder, …) — installed on demand.

/guardian listPlaybooks: recon, web_pentest, network, quick_web, osint, full_vuln
/guardian init [target]Scaffold engagement/ (authorization, scope, findings, report)
/guardian run <pb> <target>Queue a playbook checklist for the agent
/guardian toolsWhich scanners resolve on PATH / ~/.netro/agent/bin
/guardian installInstall core scanners (Windows + Linux/macOS scripts)

Windows

/guardian install
# or:
powershell -ExecutionPolicy Bypass -File "$env:USERPROFILE\.netro\agent\scripts\install-guardian-tools-windows.ps1" -CoreOnly

Linux / macOS

/guardian install
# or:
bash ~/.netro/agent/scripts/install-guardian-tools.sh
Authorized targets only. Fill engagement/authorization.md before active scans. Scanners are not bundled in the netro release — first install is the agent binary, then optional tools.

Skill: guardian-pentest (phases Planner → Tool → Analyst → Reporter, findings JSONL).

## Updates

Netro checks https://www.netrocli.bond/releases/latest.json for new versions.

/update checkCompare current vs latest + release notes
/update nowRun the platform installer (reinstall latest binary)
/update statusLast check + autoupdate flag
/autoupdate on|offStartup notify when a newer build exists
/cli-login survives updates — cli-auth.json is not cleared by the installer. You only get logged out if the operator revokes your key hash.

## Privacy / stealth

Every outbound request (web fetch, search, downloads) is rewritten to look like an ordinary Chrome browser, so target sites see normal traffic instead of an automated client. Telemetry is off by default.

/stealth                          # show status

# route ALL traffic through a proxy / Tor so the
# destination logs the proxy's IP, not yours:
export NETRO_PROXY=socks5://127.0.0.1:9050
Honest scope: your MAC address is never sent to remote servers (it stays on your LAN). Your real IP can only be hidden by routing through a proxy / VPN / Tor — that's what NETRO_PROXY does.

## Configuration

Everything lives in ~/.netro/agent/ — plain files, no magic.

~/.netro/agent/
├── settings.json     # HUD, theme, defaults
├── models.json       # providers & custom models
├── auth.json         # provider credentials (hashed)
├── skills/           # installed skills
├── prompts/          # prompt templates
├── agents/           # subagent definitions
└── sessions/         # saved sessions

settings.json

{
  "hud": {
    "display": { "showTools": true, "showTodos": true, "showCost": true },
    "colors": { "model": "cyan", "context": "green" }
  }
}

## Providers & models

Netro speaks one unified LLM API across Anthropic, OpenAI, Google, xAI, Bedrock, OpenRouter, and any OpenAI-compatible endpoint. Add a custom model in ~/.netro/agent/models.json:

{
  "providers": {
    "my-provider": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "sk-...",
      "models": [{ "id": "my-model", "contextWindow": 200000, "reasoning": true }],
      "compat": { "thinkingFormat": "openrouter", "supportsReasoningEffort": true }
    }
  }
}

Switch mid-session with /provider or /model — context is preserved.

## Skills

Skills are reusable procedures the agent can invoke. 110+ ship built-in (dev stacks, DevOps, deployment). Install more from any repo:

/skill-install github.com/acme/deploy-skills

Each skill is a SKILL.md with YAML frontmatter (name, description) plus a body of numbered steps. Skills can register their own slash commands and hot-reload per session.

## Agents

Specialist subagents run tasks in isolated context. 50+ ship built-in (verifier, silent-failure-hunter, architects, and more). Create your own:

/agent-new a reviewer that checks for security issues

Delegate to them through the subagent tool (single / parallel / chain), or let the coordinator orchestrate them for you.

## MCP

Connect Model Context Protocol servers via ~/.netro/agent/mcp.json (or project .mcp.json). Tools appear as mcp__<server>__<tool>.

{
  "mcpServers": {
    "fs": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "."]
    }
  }
}
/mcp statusConnected servers + tool count
/mcp reloadReconnect after editing mcp.json
/mcp on|offEnable / disable MCP for the session

## Cooperation / free

Netro is a local operator harness. Default posture is cooperative technical work (security testing, automation, dual-use code on systems you name as scope) without policy lectures.

/obedienceAlways-on cooperation reinforce + refusal recovery (default ON)
/freeOpt-in package: stronger free prompt, safety-gate excludes, optional soul/operator
Hosted models (e.g. some GLM / API gateways) can still refuse at the weight layer. Obedience recovers pure-text policy refusals (up to 2 follow-ups). For maximum freedom pair with local/uncensored models. This is not a jailbreak of third-party ToS — it steers the agent layer on your machine.

## Troubleshooting

Install command "308 Permanent Redirect"

Vercel redirects netrocli.bondwww.netrocli.bond (308). PowerShell irm and some curl builds don't follow 308 redirects. Use the www host exactly: https://www.netrocli.bond/install.ps1.

Windows: curl -fsSL fails in PowerShell

Use the PowerShell installer instead: irm https://www.netrocli.bond/install.ps1 | iex

Home page shows both Windows and Linux install lines

Hard-refresh (Ctrl+Shift+R). The site auto-picks your OS and shows one command; tabs switch OS without stacking both panes.

dist/dist/.../dark.json ENOENT on start

Fixed in current releases. Run /update now or reinstall from the site.

Logged out after opening netro

Your key hash was revoked or set active: false on the registry. Contact the operator for a new key, then /cli-login.

/guardian tools shows mostly missing

Scanners are optional. Run /guardian install, then open a new terminal so ~/.netro/agent/bin is on PATH.

Skill conflicts on startup

Every SKILL.md needs a description in its frontmatter. Built-in skills already have this; if you added your own, check the frontmatter.

Effort clamps to "off"

Your model doesn't advertise reasoning. Set NETRO_FORCE_EFFORT=1 (default) and add "reasoning": true to the model in models.json.

Still stuck? Reach out on Telegram: t.me/netro_1933