ZRM (Zero-dependency Router Monitor) is a lightweight, ultra-fast command-line assistant designed to map natural language onto native system operations. Operating completely offline with localized TensorFlow.js inference, it eliminates the need for heavyweight tokenization frameworks or external cloud processing APIs, returning decisions under 15ms.
- Zero External Dependencies: Bypasses sprawling processing chains (
natural, MongoDB client drivers, etc.). - Custom Unicode Regex Tokenizer: Native string splitting optimizing multi-language evaluation without allocating bulky model structures.
- Complete Privacy: 100% on-device operations. Shell behaviors, paths, and metadata remain strictly contained.
- Dynamic Command Lifecycle Metric: Time profiling tracks entire transactional cycles (Inference + Prompt Interaction + Physical OS execution times).
| Action Intent | Scope of Execution | OS Adaptation |
|---|---|---|
sys_disk |
Read available block and partition allocations. | wmic (Win32) / df (Unix) |
sys_network |
Scan sockets listening for active loopback traffic. | netstat (Win32) / ss (Unix) |
file_create |
Generates files via atomic buffers or writes deep paths. | Unified System Call |
file_rename |
Transfers metadata indices to re-assign resource naming. | Unified System Call |
file_delete |
Recursive removal protocol backed by confirmation gates. | Native File Lock Safe |
process_kill |
Drops target sockets or app sequences (Port, PID, Name). | taskkill (Win32) / pkill (Unix) |
custom_command |
Unrestricted proxy execution layer into host Shell terminal. | Raw Runtime Passthrough |
To implement new training patterns or system directives, adjust the dataset.json map:
{
"labels": ["intent_name"],
"trainingData": [
{ "text": "natural language query trigger", "label": "intent_name" }
],
"actions": {
"intent_name": {
"win32": "powershell_or_cmd_sequence",
"default": "bash_or_sh_sequence"
}
}
}