GitHub - linuxrebel/MonVisor: Product for the automated deployment, discovery and management of Promethues Env.

7 min read Original article ↗

AI-augmented monitoring configuration for Prometheus, Alertmanager, and Grafana.

MonVisor scans a network, fingerprints the services it finds, lets you approve what to monitor, and generates Prometheus/Alertmanager/rules YAML — using a local RAG knowledge base and a local Ollama model. Everything runs locally; no data leaves the host and there is no API cost.

Status: free-tier CLI is functional (discovery → review → generate → reports, plus an optional web review UI and nginx config). Deploy/Grafana automation is the paid tier and not yet implemented.

Latest release: v0.1.1 · all releases

Contents

How it works

scan  ──▶  review  ──▶  generate  ──▶  (deploy: paid)
 │           │             │
 nmap     yes/no per     RAG + Ollama draft rules.yml;
 + HTTP   service        deterministic prometheus.yml
 probe

Requirements

  • Python 3.11+
  • nmap on PATH (service discovery)
  • curl and zstd (needed to install Ollama)
  • Ollama running locally, with:
    • an LLM (default gemma4:latest)
    • the embedding model nomic-embed-text:latest
  • ~20 GB free disk to install (~16–17 GB used; 35–40 GB total recommended for working room)
  • promtool (optional) for config validation

Tested on Fedora/RHEL/CentOS/Rocky (dnf) and Debian/Ubuntu (apt). The bundled installer handles the per-distro differences automatically; see the manual steps below if you prefer to install by hand.

Install

Latest release: v0.1.1 — download monvisor-0.1.1-install.tar.gz from the releases page, then:

tar xzf monvisor-0.1.1-install.tar.gz
cd monvisor-0.1.1-install
python3 install.py

The tarball bundles the installer, the wheel, this guide, and the license — no clone or further downloads needed. The rest of this section covers installing from a clone or by hand.

The fastest path is the bundled installer. It checks and installs the system prerequisites (nmap, curl, zstd, python venv), installs and starts Ollama and pulls the models, creates a virtualenv, installs MonVisor, and runs first-time setup — all in one go:

python3 scripts/install.py

It installs MonVisor from the published GitHub release wheel when available. If there is no release yet (or you're working from a clone), build the wheel first so the installer has something local to fall back to:

pip install build           # one-time, if not already present
python3 -m build            # creates dist/monvisor-*.whl
python3 scripts/install.py  # finds and installs the local dist/ wheel

Flags: --no-sudo (skip system packages), --no-models (skip pulling Ollama models), --no-init (install only), --install-ollama / --no-install-ollama (control the Ollama install prompt), --venv PATH, --wheel PATH.

To do everything by hand instead, follow the four steps below. For a fuller walkthrough (every prerequisite, plus troubleshooting), see INSTALL.md.

MonVisor is a Python package. The full install is three parts: system tools, a running Ollama with two models, then the package itself.

1. System prerequisites

# Fedora / RHEL / CentOS / Rocky
sudo dnf install python3 python3-pip nmap curl zstd

# Debian / Ubuntu
sudo apt install python3 python3-pip python3-venv nmap curl zstd

nmap must be on your PATH (scanning depends on it). curl and zstd are needed to install Ollama in the next step — the Ollama installer is zstd-compressed and fails without it. Python 3.11+ is required.

Debian/Ubuntu note: python3-venv is a metapackage that can lag the installed interpreter on minimal images, leaving venv unable to bootstrap pip ("ensurepip is not available"). If python3 -m venv fails, install the version-specific package instead — e.g. sudo apt install python3.14-venv (match the number to python3 --version). The bundled installer does this for you automatically.

2. Ollama + models (required before monvisor init)

init, ask, and generate all use a local Ollama instance, and init builds embeddings from the bundled knowledge, so the models must be pulled before you initialize.

# install Ollama: https://ollama.com/download  (or: curl -fsSL https://ollama.com/install.sh | sh)
ollama serve &                 # start the server if it isn't already running

ollama pull gemma4             # the LLM (default model)
ollama pull nomic-embed-text   # the embedding model — REQUIRED, init fails without it

Ollama listens on http://localhost:11434 by default. Point MonVisor elsewhere with MONVISOR_OLLAMA_URL or monvisor config set ollama-url <url>.

3. Install MonVisor

Use a virtual environment to keep dependencies isolated:

python3 -m venv ~/.venvs/monvisor
source ~/.venvs/monvisor/bin/activate

# from a release wheel (recommended)
pip install monvisor-0.1.1-py3-none-any.whl

# or from the sdist
pip install monvisor-0.1.1.tar.gz

# or from a source clone
pip install .

# or editable, for development
pip install -e .

This installs the monvisor command. The knowledge base (RAG corpus + reference configs) ships inside the package, so there is nothing else to download — monvisor init is fully self-contained. (Developers can override the corpus with MONVISOR_CORPUS / MONVISOR_EXEMPLARS.)

4. Initialize

This creates ~/.monvisor/ (database, vector store, reports, configs), loads the bundled knowledge into the RAG store (embedding takes a minute), and sets a password for the web UI. Re-running init cleanly reloads the knowledge; monvisor init --reset-knowledge forces a fresh rebuild.

Verify it worked:

monvisor knowledge status       # should report 231 pairs loaded
monvisor ask "How do I run a scan?"

To leave the activated environment, run deactivate (or just close the terminal).

To run monvisor from any shell without activating the venv each time, add an alias to ~/.bashrc (or ~/.zshrc):

alias monvisor="$HOME/.venvs/monvisor/bin/monvisor"

This is preferable to auto-activating the venv on login, which would also change which python/pip the rest of your shell uses. See INSTALL.md for details.

Quick start

After monvisor init (see Install above):

monvisor env add prod 192.168.1.0/24   # register a network as 'prod'
monvisor scan prod                     # discover + fingerprint services
monvisor review prod                   # approve services to monitor (y/b/n/s/a/q)
monvisor generate prod                 # build prometheus.yml + rules.yml
monvisor ui prod                       # optional web review UI (localhost:7373)
monvisor nginx prod                    # optional reverse-proxy config

Outputs are written under ~/.monvisor/ (reports/, configs/).

Asking the AI

MonVisor has a built-in assistant that answers from its local knowledge base — both how to use MonVisor and Prometheus/Alertmanager/Grafana questions:

monvisor ask "How do I run a scan?"
monvisor ask "show me a node_exporter scrape config"
monvisor ask "How do I report a bug?"

Answers come only from the bundled knowledge (no internet). If a question isn't covered, MonVisor says so and points you to the issue tracker rather than guessing. Add --show-sources to see which knowledge snippets were used.

Configuration

Env var Purpose Default
MONVISOR_HOME Data directory ~/.monvisor
MONVISOR_OLLAMA_URL Ollama endpoint http://localhost:11434
MONVISOR_PORT Web UI port 7373
MONVISOR_CORPUS Override RAG corpus path (dev) bundled
MONVISOR_EXEMPLARS Override exemplar dir (dev) bundled

Tiers

Free (CLI): CIDR scan + fingerprinting, terminal + HTML reports, web review UI, Prometheus/Alertmanager/rules generation, blackbox fallback for un-instrumentable hosts, exporter recommendations, RAG knowledge updates.

Professional (planned): SSH config deploy, Grafana dashboard provisioning, AI-generated dashboards, Grafana-native review UI, enterprise auth (LDAP, Duo, AWS IAM, SAML).

Planning and Roadmap

The docs/ and MD-Files/ directories contain the full architecture and execution planning documents for MonVisor's next development phases (4.7–4.12).

Document Description
Architectural roadmap   [MD] [Docx] Product philosophy, deployment model, entry points, and full phase-by-phase plan for phases 4.7–4.12
Engineering map   [MD] [Docx] Technology stack, file structure, SQLite schema, ChromaDB collections, all workflows, and build phase status
Execution plan ↗ Interactive dependency tree — red = to do, green = complete. Click to open rendered in your browser.

The execution plan also opens locally from a clone (no server needed):

xdg-open MD-Files/Monvisor_execution_plan.html   # Linux
open MD-Files/Monvisor_execution_plan.html        # macOS

Development

pip install -e .
python -m build        # build sdist + wheel into dist/

License

MonVisor (the application) is licensed under the GNU General Public License v3.0 or later (GPL-3.0-or-later) — see LICENSE.

The bundled knowledge base (monvisor/knowledge/) comes from the separate MonVisor-Corpus project and is licensed CC BY-SA 4.0. The two licenses apply to their respective parts.