avai logo avai host telemetry Run it
v0.3.3 — open source, MIT licensed

Is anything shady
running on your computer?

avai is a tiny security guard for your computer. It quietly checks the places malware likes to hide, then has an AI security expert look at what it found and tell you — in plain English — whether anything is dangerous.

Think of it as a health check‑up for your laptop or server. Nothing leaves your machine.

$ docker run -p 8765:8765 -v "$PWD":/data iklob1/avai
or read the source →
26
places it checks
17
expert databases consulted
every 5 min
automatic re‑check
$0
free & open source
The problem

You can't see what your computer is really doing.

Every app you install, every browser extension, every "free" download can leave something behind — a program that starts up secretly, an add‑on that reads your passwords, a connection to a stranger's server. Your computer doesn't warn you, and there's no simple way to ask "hey, is this normal?"

Big‑company security software is expensive, complicated, and sends your private data off to someone else's cloud. Most people just hope nothing's wrong.

The fix

A plain‑English answer you can trust.

avai looks around your computer every few minutes, cross‑checks anything new against well‑known security databases, and asks an AI security expert to weigh in. You get a simple, colour‑coded list:

🔴 dangerous 🟡 worth a look ⚪ not sure 🟢 all good

Each warning comes with one clear sentence on why it's flagged and exactly what to do about it. And it all happens on your own machine — your files never get uploaded anywhere.

How it works

Three steps, then it runs itself.

🔎
1 · It looks around

avai checks the spots malware likes to hide — programs that auto‑start, browser add‑ons, network connections, plugged‑in USB devices, important system files, app privacy permissions, and live authentication events. It only reads; it never changes anything.

🧠
2 · An AI expert reviews it

Anything new gets cross‑checked against trusted security databases (the same ones professionals use), then handed to an AI security analyst that decides how worried you should be — and explains why, in one sentence. The same artifact is never analysed twice.

📋
3 · You get a simple list

Open one web page and see everything colour‑coded red / yellow / green. Filter, sort, and search every table. It re‑checks every few minutes and chimes if something dangerous shows up.

You run it once with a single command (below). After that it works on its own in the background — no accounts to create, no monthly fee, and your data stays on your computer.

What it checks

All the places trouble hides.

You don't need to know what any of these mean — avai watches them so you don't have to.

🚀 Programs that start by themselves

Malware loves to set itself to launch every time you turn the computer on. avai lists everything configured to auto‑start and flags the ones that don't belong.

🌐 Who your computer is talking to

Every app that's connected to the internet, and where. If something is quietly phoning home to a known‑bad server, you'll see it — with traffic volume and geo‑location of the remote address.

🧩 Browser extensions

That "free downloader" add‑on that can read every page and your saved passwords? One of the most common ways people get hacked. avai checks what each one can do.

🔒 App privacy permissions

Which apps have access to your camera, microphone, location, screen recording, or full disk? avai reads the macOS privacy database (TCC) and flags unexpected grants — the kind a spy tool quietly requests.

🔌 USB & Bluetooth devices

Plugged‑in drives, keyboards, and paired Bluetooth gadgets — including the kind of fake "keyboard" attackers use to sneak in commands.

📄 Tampered system files

The sensitive files that control logins and permissions — SSH keys, sudoers rules, /etc/hosts redirects. avai notices the instant one of them is quietly changed.

🛡️ Your security settings

Disk encryption, the firewall, Gatekeeper, SSH remote login, Screen Sharing, and Remote Desktop — with correct live detection on macOS. If something switched one off, that's the first thing you'll see.

🔑 Authentication events

Logins, sudo usage, Gatekeeper checks, TCC access grants — aggregated by pattern so thousands of repetitive log lines collapse into a handful of meaningful rows, each AI-classified for severity.

…and 18 more checks

Running processes, open ports, DNS lookups, installed apps, scheduled tasks, kernel extensions, quarantined downloads, MDM profiles, Wi‑Fi security, drive mounts, setuid files — 26 checks in all. If it's a place attackers hide, avai is already looking there.

Features

Everything avai watches — and does.

26 collectors on macOS (21 on Linux), 17 threat‑intel sources, and a Claude‑class model that turns it all into plain‑English verdicts.

⚙️ Processes & execution
processes process_exec_events
🌐 Network
network_connections listening_ports network_flows dns_queries network_interfaces wifi_state
📌 Persistence
launch_items kernel_extensions system_extensions mdm_profiles installed_apps
🔑 Access & identity
auth_events ssh_authorized_keys tcc_permissions privilege_config setuid_files
🛡️ Integrity & posture
system_integrity file_integrity hosts_file quarantine_events mounts
🔌 Hardware & browser
usb_devices bluetooth_devices browser_extensions
🛰️ 17 threat‑intel sources behind every verdict

Hashes, IPs, domains, URLs, CVEs, packages and OS versions are enriched before the model ever sees them.

VirusTotal MalwareBazaar URLhaus ThreatFox Feodo Tracker AbuseIPDB GreyNoise Shodan InternetDB CISA KEV NVD OSV GitHub Advisory CIRCL hashlookup crt.sh PhishTank Safe Browsing endoflife.date
🧠 AI verdicts

Every new finding is labelled malicious / suspicious / unknown / benign — with a MITRE category, a confidence, and a one‑line fix. Bring your own key (Anthropic or any litellm provider).

One docker run

Same image is both dashboard and monitor.

No SIEM · no agent · no cloud

Runs entirely on your host.

Dedup by content hash

The same artifact is never judged twice.

Read‑only dashboard

Search, filter, sort, paginate, audio alerts.

Just a SQLite file

Point the dashboard at any avai.db.

macOS & Linux

One tool, both platforms.

Native install

pip install avai-monitor.

Open source · MIT

Auditable, model‑agnostic.

Why avai

The pros, in plain terms.

What you get for one docker run.

✅ Answers, not logs

Every finding comes back in plain English with a verdict, a confidence, a MITRE category, and a concrete fix — no query language, no triage spreadsheet.

📦 Zero infrastructure

One container. No SIEM, no agents to enroll, no control plane to run or pay for.

🔐 Private by default

Everything runs on your machine; you bring your own model key, and only new findings ever leave — for a lookup or the LLM call you opted into.

💸 Cheap to run

Content‑hash dedup judges each artifact once — a busy host doesn't mean a big bill — and cached intel hits skip the network entirely.

🔭 EDR breadth, no agent

26 host surfaces × 17 intel sources behind a single verdict — the coverage of an endpoint product without anything to install on every machine.

🖥️ Cross‑platform

macOS and Linux from the same tool, with platform‑correct live detection.

🧰 Open and yours

MIT‑licensed, auditable, and model‑agnostic — swap to any litellm provider with one env var.

🟢 Safe for production

Collectors only read; the dashboard is read‑only. Point it at a server without touching it.

📤 Portable history

The whole state is a single SQLite file — scan on a server, view on your laptop, archive a snapshot, diff over time.

The AI expert

Like having a security analyst on call 24/7.

Finding things is easy; knowing which ones matter is the hard part. avai hands each new discovery to an AI security analyst (powered by Claude) that explains, in one plain sentence, why it's safe or risky — and tells you exactly what to do next.

  • It cross‑checks against 17 trusted security databases before deciding.
  • It only reviews what's new — the same finding is never sent twice.
  • Every alert says why and how to fix it — no security degree required.
  • Resolved problems disappear from the list automatically.
  • Auth events are classified by recurring pattern, not individual log lines — so noisy housekeeping is marked benign once and never surfaced again.

Here's what real alerts look like — three findings from one check:

Auto‑start program 🔴 dangerous
A hidden "updater" that runs at every login

It launches a program from a temporary folder every time you sign in, and it isn't signed by a known software maker — a classic way malware stays on a machine.

→ Remove it, then check that temp folder for what it was running.

Browser extension · Chrome 🟡 worth a look
"Free YouTube Video Downloader"

For a video downloader, it asks for a lot — it can read every website you visit and your login cookies. That mismatch is a common sign of a sketchy add‑on.

→ Turn it off and check when it was installed.

Internet connection 🟢 all good
Spotify talking to Spotify's servers

A trusted, signed app making a normal, encrypted connection to its own company. Nothing to worry about.

What you see

One simple web page.

Open it in your browser and everything's right there — colour‑coded findings, live network flows, DNS queries, open ports, auth events, privacy permissions and more. Every table has search, filters, and pagination. It refreshes itself and chimes if something dangerous appears.

localhost:8765 — avai dashboard
avai dashboard overview: run and collector counts, verdict-totals donut, macOS system integrity, collector errors, a 12-hour verdict chart, and the live findings table
The overview — run & collector counts, the verdict‑totals donut, macOS system‑integrity toggles, collector errors, a 12‑hour verdict chart, and the live findings table.
A single finding expanded showing the LLM reasoning, remediation, and the raw collected process data
Finding detail. Expand any row for the LLM's reasoning, a one‑line remediation, and the raw collected data (pid, cmdline, user, content hash) behind the verdict.
Network flows aggregated by destination, with an IPv6 high-port connection flagged as a suspicious C2 beacon
Network flows. The tcpdump aggregator flags an IPv6 high‑port C2‑beacon candidate as suspicious while CDN, mDNS and LAN traffic stay benign.
Per-destination network flows enriched with the owning process, ASN, geo, traffic volume, and a rationale
Enriched flows. Every destination tied to its owning process, ASN/geo, traffic volume, and a one‑line “why”.
Findings table with filters, plus rows-per-collector volume and recent-run history
Findings, collectors & runs. Filterable findings, plus how much each collector pulled and the recent‑run history.
The same dashboard against a different host and cycle, with 3,426 verdicts and suspicious AirWatch persistence
The same dashboard, another host — a different cycle with 3,426 verdicts and suspicious AirWatch/MDM persistence surfaced for review.
Get started

One command to run it.

This part is for whoever sets up your computer or server — it's a single line they paste into a terminal. Not sure what that means? Send them this page; they'll know exactly what to do. Once it's running, you just open the web page.

Docker
recommended

Dashboard runs anywhere. Monitor needs a Linux host.

docker run -p 8765:8765 \
  -v "$PWD":/data \
  iklob1/avai
docker compose

Monitor + dashboard, both auto-restart.

curl -O https://raw.githubusercontent.com/\
iklobato/avai/main/docker-compose.yml
docker compose up -d
pip (macOS or Linux)

Full visibility — including TCC, Gatekeeper, auth events — with sudo on macOS or Linux.

pip install 'avai-monitor[judge]'
sudo avai monitor &
avai dashboard
Common recipes
🔑 Turn on the AI verdicts

Add an API key; everything else stays the same.

docker run -p 8765:8765 -v "$PWD":/data \
  -e ANTHROPIC_API_KEY=sk-ant-... \
  iklob1/avai avai monitor --db /data/avai.db
🖥️ Watch a Linux server, keep it running

Install once; it re-checks every 5 minutes on its own.

pip install 'avai-monitor[judge]'
sudo -E avai monitor --db /var/lib/avai/avai.db --interval 300 &
avai dashboard --db /var/lib/avai/avai.db --host 0.0.0.0
🍎 Full macOS coverage (TCC, Gatekeeper, auth events)

Native install with sudo for complete visibility — including privacy permissions and system integrity.

pip install 'avai-monitor[judge]'
export ANTHROPIC_API_KEY=sk-ant-...
sudo -E avai monitor &
avai dashboard
📟 See the alerts without opening a browser

It's one file — ask it for the things worth worrying about.

sqlite3 -box avai.db "SELECT verdict, collector, reasoning
  FROM judgements
  WHERE verdict IN ('malicious','suspicious');"
👀 Already have a scan? Just view it

Point the dashboard at any avai.db, on any computer.

docker run --rm -p 8765:8765 -v "$PWD":/data iklob1/avai

Full guide & more examples in the README.

Everything's MIT-licensed and lives at github.com/iklobato/avai.