vex

IOC enrichment hub for SOC triage and DFIR — VirusTotal + AbuseIPDB / Shodan / WHOIS / MISP / OpenCTI, straight from your terminal.

 ██╗   ██╗███████╗██╗  ██╗
 ██║   ██║██╔════╝╚██╗██╔╝
 ██║   ██║█████╗   ╚███╔╝
 ╚██╗ ██╔╝██╔══╝   ██╔██╗
  ╚████╔╝ ███████╗██╔╝ ██╗
   ╚═══╝  ╚══════╝╚═╝  ╚═╝

vex takes an indicator (a hash, IP, domain, or URL) and turns it into a verdict with context. It started as a VirusTotal CLI and grew into an enrichment hub: one primary source (VirusTotal) plus optional secondary sources that add reputation and your own threat-intel. It runs standalone, and it slots into a pipeline (barb → vex → sift) over stdin/stdout.

Install

pip install vex-ioc          # core
pip install vex-ioc[ai]      # + Anthropic / OpenAI explanations (Ollama needs no extra)

Quickstart

vex config --set-virustotal YOUR_VT_KEY  # or set VT_API_KEY
vex triage 8.8.8.8
vex triage evil.com --explain            # add an AI narrative
vex investigate <sha256> -o rich         # deep dive

Secondary enrichers (optional, saved to ~/.vex/config.yaml):

vex config --set-abuseipdb YOUR_KEY --set-shodan YOUR_KEY
vex config --set-misp-url URL --set-misp-key YOUR_KEY
vex config --set-opencti-url URL --set-opencti-token YOUR_TOKEN

IOC: 8.8.8.8 (ipv4)
Verdict: CLEAN
Detections: 0 malicious / 0 suspicious / 94 engines
Reputation: 0

What you get

• Two modes — triage (fast verdict, minimal API calls) and investigate (deep DFIR: PE info, sandbox behavior, passive DNS, relationships, MITRE ATT&CK, timeline).
• Multi-source enrichment — VirusTotal primary; AbuseIPDB, Shodan, WHOIS, MISP, and OpenCTI as secondary enrichers (key-gated, fail-open).
• Batch correlation — --correlate clusters a batch of IOCs by shared infrastructure (ASN, malware family, contacted IPs/domains).
• AI explanations — opt-in --explain narratives (Claude / OpenAI / local Ollama / deterministic template), with prompt-injection defense on the untrusted data fed to the model.
• Pipeline-ready output — JSON, NDJSON (streaming), CSV, STIX 2.1, ATT&CK Navigator, self-contained HTML; verdict-mapped exit codes.
• Quota-aware batches — IOC dedup, an up-front ETA, a --max-quota budget guard, and a persistent daily VT-quota counter with a stderr status line (on main, pending 1.7.0).
• Local knowledge base — vex tag / note / watchlist annotate IOCs in ~/.vex/. Re-triage an entire watchlist in one shot with vex watchlist run <name> (on main, pending 1.7.0).
• TI write-back (on main, pending 1.7.0) — vex investigate --sight writes MISP sightings and OpenCTI observables for IOCs above a configurable verdict floor; triple opt-in (config killswitch + flag + dry-run preview).

Verdicts

Enrichment sources

VirusTotal is the primary source; it produces the verdict. The others are secondary enrichers that augment an investigate result.

Correlation

vex triage -f iocs.txt --correlate            # deterministic cluster table
vex triage -f iocs.txt --correlate --explain  # + per-cluster AI campaign narrative

--correlate groups a batch by shared ASN, malware family, contacted IPs, and contacted domains — surfacing likely campaigns. It is deterministic by default; the AI narrative is added only when --explain is also set.

AI explanations

--explain adds a narrative and next steps; --explain-model overrides the model. Providers: Anthropic, OpenAI, local Ollama, and a deterministic template fallback when none is configured.

export VEX_AI_PROVIDER=ollama        # local, no cloud key
vex investigate evil.com --explain

Output formats

-o rich (default TTY) · -o console · -o json · -o ndjson (one result per line, streamed) · --csv · --stix · --navigator (ATT&CK layer, investigate) · --html <path> (self-contained report). All formats show real IOCs by default; --defang defangs them (e.g. evil[.]com); HTML reports always defang.

sift triage alerts.json -o json | vex triage --from-sift -o ndjson

Pipeline

flowchart LR
    barb -->|URLs / IOCs| vex
    vex -->|enriched IOCs| sift
    sift -.->|extracted IOCs| vex

vex is the enrichment hub. --from-barb reads barb JSON; --from-sift reads a sift TriageReport, extracts its IOCs, and enriches them — closing the sift ↔ vex loop. NDJSON output and verdict exit codes make it scriptable downstream.

Scale

For large batches, such as IOCs extracted from a SIEM export:

• Dedup — duplicate IOCs are collapsed before any query, which saves quota; --no-dedup disables it.
• ETA + counters — an up-front estimate and a processed (N from API, M cached) summary on stderr.
• --max-quota N — caps fresh API lookups; cached IOCs are still served, the rest are skipped with a notice. Re-run to continue — the SQLite cache resumes where you left off.

Configuration

Priority: --api-key flag → environment variable → ~/.vex/config.yaml → defaults.

vex config --show        # active config, masked secrets
vex doctor               # what's configured (no network)
vex doctor --probe       # + live connectivity, surfaces the actual error

enrichment.stix_tlp_version (1.0 default, 2.0) selects which STIX TLP marking ids the export emits.

TI write-back (on main — pending 1.7.0)

vex investigate --sight writes MISP sightings and OpenCTI observables back for IOCs that meet a configurable verdict floor. Three gates must all be open before any write occurs:

1. enrichment.writeback_enabled: true in ~/.vex/config.yaml (default false).
2. --sight flag on vex investigate.
3. Use --dry-run-sight first to preview the payloads without sending.

vex investigate 1.2.3.4 --dry-run-sight     # preview payloads, no network
vex investigate 1.2.3.4 --sight             # write if enabled and above floor
vex investigate -f iocs.txt --sight         # batch write-back

Key config keys (in enrichment: section):

Security

Exit codes

Computed from the highest-severity verdict in the run:

Known limitations

• MISP (v2.5.38) and OpenCTI (demo v7.26) lookups are live-verified end-to-end, and both have mock tests. They query real instances and stay version-sensitive: if a server's schema differs, the enricher fails open (no enrichment, no crash). Use vex doctor --probe to confirm connectivity. pycti is a possible future fallback if OpenCTI's GraphQL schema drifts.
• AI: Anthropic and Ollama are live-verified end-to-end; OpenAI shares the same code path (mock-tested). Without a provider, --explain falls back to a deterministic template.
• Secondary enrichers run on investigate, not triage. A triage --explain gets no AbuseIPDB / Shodan / MISP / OpenCTI context by design.
• STIX TLP:CLEAR maps to the TLP 1.0 WHITE marking id by default — set enrichment.stix_tlp_version: "2.0" to switch to TLP 2.0 ids.
• The VirusTotal free tier allows roughly 4 requests/min, so large batches are quota-bound — see Scale.

Docs

Full manual: docs/ — getting started, every command/flag, the enrichment model, output formats, the barb/sift pipeline contracts, and configuration. Built-in guide: vex manual (and vex manual ai / pipeline / config). Changelog: CHANGELOG.md.

Author

Christian Huhn — building security tooling for SOC/DFIR workflows.

• GitHub: @duathron
• LinkedIn: Christian Huhn

License

MIT