NetClaw – AI‑powered network‑operations assistant

What it is – NetClaw is a locally‑run, LLM‑backed “co‑pilot” for network engineers. It sits on top of an existing AI‑agent runtime (OpenClaw or Hermes) and adds a large catalog of skills (237) and MCP integrations (176) that let it read network documentation, query live devices, propose and audit configuration changes, and keep a tamper‑proof audit trail.

How you use it

• Installation – A Bash‑based interactive installer (scripts/install.sh) lets you pick a profile (recommended, Cisco‑only, multivendor, cloud, security, labs, etc.) or hand‑pick individual components. The installer can also run non‑interactively with --profile, --components or --all flags. It configures the chosen runtime, creates a .env file, and registers the selected MCP servers.
• Runtime choice – By default NetClaw runs on OpenClaw; you can switch to Hermes (Nous Research’s self‑improving agent) with --runtime hermes. Both expose the same shared HUD UI.
• HUD (visual workbench) – A browser UI (:3000 → HUD API :3001) provides a chat pane, a Canvas for branching investigations, and panels for RAG, token usage, security mode, and specialist “Jev” views. The HUD works on macOS, Ubuntu (WSL2) and Linux.
• Interaction channels – Chat from the HUD, VS Code extension, mobile Flutter client (iOS/Android), or a Zoom Realtime Media Streams app. Each channel respects the same approval and audit policies.
• RAG (retrieval‑augmented generation) – Upload PDFs, Markdown, HTML, Office docs, or URLs. NetClaw builds a local semantic‑keyword index, reranks results, and returns cited passages with document, section and page numbers. Knowledge stays separate from the LLM’s memory.
• GCF (compact network evidence) – A serializer that compresses tool output (tables, topology graphs, deltas) into a smaller token representation, reducing context size for LLM calls.
• Mobile & Zoom – A Flutter app lets engineers approve changes from a phone; a Zoom side‑panel can trigger investigations from meeting transcripts.
• VS Code extension – Adds a NetClaw activity bar inside VS Code for chat, RAG browsing, test‑bed inventory editing, MCP configuration, and visual BGP topology.

Key capabilities

• Live network investigation – Query devices, run nmap, pull BGP state, compare against uploaded design guides.
• Change automation with safeguards – Propose configuration edits, run them through a baseline, verify results, and record an immutable audit trail.
• Specialist “NetClaws” – Deploy multiple focused agents behind a single Border for risk‑segmented tasks.
• Extensible skill catalog – 237 pre‑written skills (e.g., pyATS, NetBox, ServiceNow, AWS, Azure, FortiManager, etc.) that can be enabled/disabled per installation.
• Security & compliance – All actions require explicit approval; audit logs, token‑omics, and a “Security mode” panel make it clear what the LLM is allowed to do.

Who it’s for – Network operations teams, NOC engineers, and security analysts who want an LLM‑driven assistant that works inside their own environment, never sending raw network data to the cloud, and that integrates with the tooling they already use.

Where to start – Clone the repo, run ./scripts/install.sh (choose a profile), pick OpenClaw or Hermes, and open http://localhost:3000 to launch the HUD. The README links to detailed docs for each piece (RAG pipeline, GCF serializer, mobile build, Zoom integration, VS Code extension, and safety specifications).