A Deterministic Agentic Engine and CLI for Exposure Management.

Built on Tenable. Grounded in your data. Reproducible by design.

Wiki · DeepWiki · Remote Execution · Email Setup

navi turns your Tenable exposure data into something an AI agent — or an engineer at a terminal — can act on with confidence. It syncs Tenable Vulnerability Management, Tenable One, and Tenable WAS into a local, queryable database, exposes every capability as a real tool, and ships a pack of expert agents (the Hounds) that hunt specific classes of exposure. No dashboards to click through, no data left on the table, and — crucially — nothing made up.

Most AI agents guess. They summarize, approximate, and occasionally invent an asset or a CVE that was never there. That's fine for a chat. It is unacceptable for security work, where a wrong answer means an unpatched host or a bad decision made with confidence.

navi is a Deterministic Agentic Engine: every answer an agent gives and every action it takes is a real, reproducible tool call — against a local mirror of your Tenable data (navi.db) or the Tenable API directly. Ask the same question twice and you get the same answer. There is no hallucinated finding, no fabricated asset, no fuzzy summary standing in for ground truth.

The intelligence lives in the orchestration — deciding what to hunt, how to tag, when to scan, what to remediate. The facts always come from your data. That separation is the whole point: agentic flexibility on top of deterministic truth.

flowchart LR
    T[Tenable VM / One / WAS] -->|sync| N[navi engine]
    N <-->|SQL| DB[(navi.db)]
    N -->|API writes| T
    subgraph Agent side
      C[Claude / any MCP client] --> M[navi-mcp server]
      SK[Skills] -.guides.-> C
      HO[The Hounds] -.playbooks.-> C
    end
    M -->|deterministic tool calls| N

navi is four things that reinforce each other:

• The engine (CLI). navi pulls Tenable data into navi.db, a local SQLite mirror you can query in milliseconds, and turns tagging, ACR calibration, scanning, exporting, and remediation into single commands. This is the deterministic core everything else drives.
• The MCP server. navi-mcp exposes the engine to any MCP-capable AI (Claude and others) as typed tools, behind a write-gate so read operations are always safe and state-changing actions require an explicit opt-in.
• The Skills. Packaged best-practice playbooks that teach an agent how to drive navi well — the right selectors, the safe patterns, the gotchas — so the model orchestrates instead of improvises.
• The Hounds. A pack of exposure-management agents, each trained on one kind of hunt — known-exploited vulnerabilities, expiring certificates, attack paths, unsupported software, and more — grounded in exact navi selectors rather than generic advice.

Each Hound is a focused exposure hunter with its own detection logic, tag scheme, and navi selectors. Run one by name, or release the whole pack.

navi needs Python 3.9+ (3.12 recommended) and a Tenable API key pair.

# 1. Install
pip install navi-pro

# 2. Add your Tenable API keys (stored locally, never echoed)
navi config keys --a <ACCESS_KEY> --s <SECRET_KEY>

# 3. First sync — pulls your Tenable data into navi.db.
#    This is long-running on large tenants; run it on the CLI, not through MCP.
navi config update full

Then start exploring and acting:

# See what navi sees
navi explore info assets
navi explore data query "select ip_address, fqdn from assets limit 10;"

# Tag every asset with a CISA KEV finding
navi enrich tag --cisa

# Export risk by business segment (includes ACR + AES)
navi export bytag

# Calibrate criticality for your crown jewels
navi enrich acr --tag "Tier:Production" --mod set --acr 9 --reason business

Prefer containers? A Dockerfile for the full navi + navi-mcp image lives under navi-mcp-suite/ — Ubuntu 24.04, Python 3.12, isolated venv, navi.db on a mounted volume.

Give any MCP-capable AI safe, deterministic access to navi:

# Build the image (from the repo root)
docker build -f navi-mcp-suite/Dockerfile -t navi-mcp:latest .

# Store keys + do the first sync into a persistent volume
docker run --rm -v navi-data:/data navi-mcp:latest navi config keys --a <ACCESS> --s <SECRET>
docker run --rm -v navi-data:/data navi-mcp:latest navi config update full

# Serve over HTTP on :8000
docker run -d -p 8000:8000 -v navi-data:/data navi-mcp:latest

Core runtime — where the server finds navi and its data:

Capability gates — every one defaults to off; set to 1 to unlock the tools it guards:

Integration keys — for the agent and delivery layers:

Tenable API keys are not environment variables — set them once into the mounted volume with navi config keys --a <ACCESS> --s <SECRET> so they live in navi.db, never in the process environment or an image layer.

Reads are always safe. Anything that changes state is double-guarded: the server must be started with NAVI_MCP_ALLOW_WRITES=1, and each write call must pass confirm=True. The two most sensitive actions raise the bar further — emailing (NAVI_EMAIL=1) and remote command execution (NAVI_REMOTE_CODE_EXECUTION=1) each need their own dedicated flag on top of the write-gate. An agent can look all day; it can only act when you've explicitly let it.

Skills are the connective tissue between an AI and the engine — they encode the correct way to use each part of navi so the model orchestrates expert workflows instead of guessing at flags.

Tenable Vulnerability Management (Tenable.io), Tenable One / Exposure Management, and Tenable WAS. navi is battle-tested at scale — it has ingested tenants with 1.5M+ assets and 160M+ vulnerabilities into navi.db, with plugin queries returning in seconds. FedRAMP tenants are supported via URL configuration.

• Wiki — full command reference and guides
• DeepWiki — ask questions about the codebase
• PUSH.md — remote command & file execution over SSH
• RESEND_SETUP.md — one-time email setup

Troubleshooting (Docker memory, DB locks, key checks, scale tuning) now lives in the Wiki so this page stays focused.

navi is released under the MIT License. See LICENSE.