You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Run assessments only on infrastructure you own or have written authorization to test.
What it is
AIDOS is an LLM-orchestrated denial-of-service resilience assessment agent. It uses Gemini as a reasoning engine to autonomously conduct multi-phase availability assessments: reconnaissance, fingerprinting, analysis, attack execution, impact measurement, and report generation β without human input between steps.
The research motivation is whether an LLM agent can replicate the decision-making process of a human security engineer performing availability testing: selecting attack vectors based on observed target characteristics, adapting strategy based on results, and compounding multiple simultaneous vectors.
Tools (49)
Recon
Tool
Description
detect_installed_tools
Inventories available system tools
crawl_endpoints
Spiders the site, collects URLs, forms, response times
detect_tech
Fingerprints server, CDN, WAF, framework, language from headers
Enumerates real origin IP behind CDN via subfinder, crt.sh, DNS A/MX/TXT/SPF records, and common subdomain patterns (direct., origin., mail., staging., etc.). Verifies each candidate by connecting with the correct Host header and confirming a valid HTTP response.
Analysis
Tool
Description
http_request
Single HTTP request, full response inspection
benchmark_endpoint
N sequential requests, returns avg/p95/p99 latency
test_rate_limit
Rapid sequential requests, detects 429 and records block threshold
Tests CDN cache bypass techniques: random param, range header, pragma, XFF variants
redos_probe
Sends catastrophic backtracking regex payloads, measures response time change
find_amplification_ratio
Scans all discovered endpoints for response/request size ratio. Identifies highest-cost flood targets β an endpoint returning 500KB per 100-byte request is more efficient to flood than one returning 200 bytes.
L7 Flood
Tool
Description
http_flood
Async HTTP flood, built-in, no external tools required
spoof_flood
HTTP flood with rotating User-Agent and X-Forwarded-For per request
flood_origin
Connects directly to origin IP with Host header set to the real domain, bypassing CDN
ipv6_prefix_flood
Each connection binds to a different source IPv6 from a routed prefix. Genuine per-connection source IP diversity at L7. Requires a /48+ IPv6 block routed to the attack machine.
http2_rapid_reset
CVE-2023-44487 built-in implementation. No external tools required. Opens N persistent HTTP/2 connections and sends HEADERS+RST_STREAM pairs continuously. Server must allocate and free stream state for each pair. Operates below application-layer rate limiters.
h2load_flood
HTTP/2 multiplexed stream flood via h2load (external). Fallback to bombardier --http2.
bombardier_load
External: bombardier
vegeta_attack
External: vegeta, constant-rate load
wrk_benchmark
External: wrk
siege_load
External: siege
k6_load
External: k6, scripted virtual user load
L4
Tool
Description
hping3_flood
SYN, UDP, ICMP, ACK, RST, XMAS, FIN packet flood via hping3
The tool operates from a single machine. Effectiveness against CDN-protected targets depends on whether origin IP disclosure vulnerabilities exist. Against properly configured CDN with no origin exposure, the primary available vector is ssl_handshake_flood at the CDN edge and application-layer attacks that the CDN forwards (GraphQL, ReDoS, xml_bomb).
License
MIT
About
Autonomous LLM-powered DoS resilience assessment framework that orchestrates open-source security tooling for recon, analysis, stress testing, and reporting.