— Open-Source AI Security Agents
Hunt Every Exposure.Keep a Human Holding the Leash.
The Hounds is an open-source pack of AI security agents that runs on your Tenable data. They find the unknown, prioritize by real exposure, and propose every change for your approval — grounded in real queries, never guesses.
Built by the author of navi — 353K+ installs. Independent and open-source.
Built on Tenable. Grounded in your data. Reproducible by design. Most AI agents guess. The Hounds don't — every answer and action is a real, reproducible tool call against your data. Ask the same question twice, get the same answer. No hallucinated finding, no fabricated asset.
Listed on the Tenable CyberAgents Exchange
Verified open-source agent pack for Tenable Exposure Management
— The Atlas Ownership Map
See the surface. Shrink the work.
Two views. One goal: turn tens of thousands of findings into a handful of owned paths.
Who owns what — at a glance.
Every route and path threaded to its owner, with the gaps called out: unowned, under-owned, and ownerless risk. From the Atlas Ownership Map.
Follow the risk to a human.
Routes and paths flow to the owners on the hook for them — and the ownerless risk stands out immediately. Nothing gets fixed until someone owns it.
— The Pack
41 agents. One pack.
KEV, attack paths, identity, certificates, post-quantum crypto, AI, IoT, software, ownership, remediation — and the orchestrator that ties them together.
See all 41 →— Why Trust the Hounds
Built for security teams who need answers, not estimates.
Agents propose. You approve.
Every tag and score change is gated behind human confirmation. Nothing runs autonomously until you arm it.
Real queries, not estimates.
Every answer comes from a real query against your local data — or the hounds tell you they can't answer.
Honest about blind spots.
Uncredentialed hosts and stale data get called out, not hidden.
— The Problem
Beat vulnerability fatigue: the mountain becomes a short, owned list.
Teams don't burn out because they can't find vulnerabilities — they burn out because there are too many, with no way to separate signal from noise.
Real output from a reference run: 413 CVE line-items → 53 unique fixes → 6 owned apps. 98.5% fewer things to track — and one Chrome patch cleared 90 CVEs in a single action.
Rank, don't dump
Correlated risk floats the critical few to the top, so the long tail stops screaming for equal attention.
Your slice only
Ownership means each person sees just their assets — a handful to act on, not the whole estate's mountain.
Kill the noise
Duplicate and ephemeral assets get cleaned up, so you stop re-triaging the same ghost host ten times over.
— The Closed Loop
From a scan result to a verified fix — and back.
Most tools stop at step 2. The Hounds run the whole loop — discover, prioritize, assign, deliver, remediate, verify, report — and every step keeps a human in control.
Every step is gated — nothing tags, emails, or changes anything in your tenant until you approve it.
— You Hold the Leash
One pack, three levels of autonomy.
Turn the dial from fully governed to fully autonomous — the same grounded, non-destructive guarantees run across all three.
Governed
the engine (repo)
Self-hosted repo. The navi CLI runs deterministic agents — every write is proposed and confirmed. Maximum control, data local, near-zero tokens.
Assisted
skills + MCP (artifact)
Claude drives inside a structured console over the MCPs — conversational, but guided by 17 skills. Great for exploration and triage.
Autonomous
navi-mcp + the harness
Prompt an outcome and Claude plans and acts across the whole pack over the navi MCP — grounded by the Hounds harness. One prompt, whole estate.
— The Payoff
Less noise. Faster fixes. Real accountability.
The whole estate's exposure, ranked — the morning read that replaces fifteen tool logins.
gets an owner and a route to a human, so nothing rots unassigned.
Each team gets only their tasks, with the fix one click away in Tenable.
Gated, logged, and verified — you can show what changed and whether it stuck.
— Open Source
Free. Open. Yours to run.
Install the navi CLI and start hunting exposures in your own environment today.
— Why not just use native agentic AI?
Open-source, not a black box.
Every query, every prompt, every decision is in the repo. Read it, fork it, audit it.
Your vulnerability data stays local.
The pack runs against your local navi database. Nothing is sent to a third-party cloud.
Grounded and gated — it tells you when it can't answer.
Queries are forced against real data. If the answer isn't there, the hound says so.