devvm

Free, open-source (MIT) command-line tool that gives each project folder its own virtual machine, built for running coding agents (Claude Code, Codex, opencode and pi) away from your real files and keys. Run devvm in a project and it detects the toolchains your files call for (a package.json gets node), builds the VM on Lima, mounts only that project folder at the same path as on your machine and opens a shell in it. New VMs start on an allowlist network tier so they reach only the hosts you approve, and your secrets, such as your GitHub and Claude Code tokens, stay on your machine. Built-in docker, nix and desktop layers; releases are signed with a minisign key. Needs macOS 13 or newer, or Linux, on arm64 or x86_64 (tested on Apple silicon Macs and on Linux, not yet on Intel Macs). About 430 GitHub stars; v0.1.0 shipped Nov 6, 2026 (v0.1.1 is still a release candidate) and its Show HN reached about 190 points. Very new, so treat it as early software.

FREE TIERchecked 11/15
Category
AI for IT & Cybersecurity
Price
Free and open source (MIT)
Free option
Yes
Last checked
Nov 15, 2026

sandboxagentsvirtual machineisolationopen sourcenew

Visit devvm ↗

Get the weekly dropThe best new AI tools, one email a week. Free.

Related AI for IT & Cybersecurity tools 62

See all
Tool
Price
Status
Link
TryHackMe (Echo)Hands-on cyber training labs with Echo, a built-in AI tutor that gives context-aware hints without handing you the answer.
Price See site
See site
HTB Academy (HTB Coach)Hack The Box's training modules with HTB Coach (beta): AI explanations, section summaries and quizzes tied to what you're studying.
Price See site
See site
PentestGPTOpen-source autonomous pentesting agent from USENIX Security 2024 research, with staged recon-to-exploit runs for CTFs and lab targets.
Price Open source (MIT)
FREEchecked 11/15
Burp AI (PortSwigger)AI pentesting partner built into Repeater in Burp Suite Professional for validating findings and cutting scan noise.
Price See site
See site
SnykDeveloper security platform that scans your code, open-source dependencies, containers and IaC from the IDE, CLI and repos.
Price Team from $25/mo (up to 10 developers)
FREE TIERchecked 11/15
SemgrepFast code scanner (SAST, supply chain, secrets) with AI-powered detection, triage and autofix. Great for scanning your own home-lab scripts. New Nov 12, 2026: Semgrep Mandoline, a research preview of program-analysis tools built for AI agents rather than people: call graphs and cross-file dataflow (taint) analysis across 50 languages that an agent can query while it hunts for vulnerabilities. Semgrep says it is 50x faster than its Pro engine for these agent-style analyses. Existing Semgrep customers get it inside Semgrep Agentic Workflows and the Context Layer; others can ask to join the research preview.
Price Free for up to 10 contributors (60 AI credits); Teams Code from $30/contributor/mo
FREE TIERchecked 11/15
GitHub Copilot AutofixAI fixes for CodeQL code scanning alerts. Agentic autofix (GA Oct 15, 2026) investigates alerts and opens a draft PR with a validated fix.
Price Free on public repos; private repos need GitHub Code Security
FREE TIERchecked 11/15
VirusTotal Code InsightAI summaries in VirusTotal that explain what a suspicious script (starting with PowerShell) or decompiled code actually does. Handy for malware-analysis practice.
Price Free for community users on eligible uploads; Code Insight API limited to 50 requests/day
FREE TIERchecked 11/15
TinesSecurity and IT workflow automation (SOAR) with built-in AI and bring-your-own-model support, used by SOC and IT teams.
Price Free edition: 3 live workflows + one-time $50 AI allowance; paid editions on request
FREE TIERchecked 11/15
SentinelOne Purple AIAgentic AI security analyst in the SentinelOne Singularity Platform: ask questions in plain English, auto-triage alerts and run guided investigations.
Price See site
See site
WazuhFree, open-source SIEM and XDR used by home labbers and SOC teams for log analysis, threat detection and compliance. Wazuh documents AI threat hunting with local models via Ollama, and Wazuh 5.0 (now in release candidate) adds a built-in AI Assistant. Wazuh 4.14.10 (Nov 9, 2026) is a maintenance release for 4.x with security fixes, including path traversal in the agent diff-folder cleanup and a root-level OS command injection in the rootcheck promiscuous-interface check, so home labs on 4.x should plan the upgrade. About 17,600 GitHub stars.
Price Free and open source (self-hosted); Wazuh Cloud is a paid hosted option
FREE TIERchecked 11/15
CyberStrikeAIOpen-source AI pentesting platform written in Go: agents run 100+ security tools over MCP, with human approval steps, audit logs and CTF roles. Only for labs and systems you're authorized to test.
Price Open source (Apache-2.0)
FREEchecked 11/15
Sophos AI DefenseFinds the AI tools employees and apps are using, including unapproved shadow AI, and enforces real-time guardrails on prompts and data across endpoint, network, identity and browser. Generally available Oct 16, 2026.
Price Quote-based add-on for Sophos EDR, XDR and MDR customers
checked 11/15
garak (NVIDIA)NVIDIA's open-source LLM vulnerability scanner: probes chatbots and AI apps for prompt injection, jailbreaks, data leakage, hallucination and toxicity, then reports what got through. v0.17.1 shipped Oct 16, 2026.
Price Open source (Apache-2.0)
FREEchecked 11/15
AteraAll-in-one IT management (RMM, help desk, ticketing and patching) priced per technician, with AI Copilot built into every plan for scripts, ticket summaries and knowledge-base drafts. Robin, its end-user AI agent, is sold separately.
Price Professional from $149/technician/mo billed annually ($169 monthly); 30-day free trial
checked 11/15
Security OnionFree and open platform for threat hunting, network security monitoring and log management (2+ million downloads), a staple of SOC home labs. Security Onion 3 shipped in 2026. Onion AI, its assistant for alert analysis and detection tuning (now with local model support), comes with Security Onion Pro. New in 3.4.0 (Nov 2, 2026): Agentic AI Automations for Pro users, plus updated Suricata 8.0.8, Zeek 8.0.11 and Elasticsearch 9.4.8. Security Onion 2.4 reached end of life on Oct 1, 2026.
Price Free and open; Onion AI is part of Security Onion Pro (contact sales)
FREE TIERchecked 11/15
promptfooOpen-source CLI for testing and red-teaming LLM apps and agents: write evals in YAML, compare models side by side and scan for prompt injection, jailbreaks and data leaks. About 26,000 GitHub stars; the site says 300,000+ developers use it.
Price Community edition free and open source (MIT), red teaming up to 10k probes/mo; Enterprise custom
FREE TIERchecked 11/15
Nuclei (ProjectDiscovery)Fast open-source vulnerability scanner built on thousands of community YAML templates, with an -ai option that generates and runs a template from a plain-English prompt. About 32,000 GitHub stars. Only scan systems you own or are authorized to test.
Price Open source (MIT); free AI template editor on ProjectDiscovery Cloud
FREEchecked 11/15
FreshserviceCloud IT service desk (ITSM) from Freshworks: ticket and incident queues, knowledge base, self-service portal and service catalog, with Freddy AI Copilot as an add-on. A common help desk platform you can practice on with the free trial.
Price Starter from $19/agent/mo billed annually; Freddy AI Copilot add-on $29/agent/mo; free trial
checked 11/15
Microsoft Security CopilotGenerative AI for security teams across Microsoft Defender, Sentinel, Entra, Intune and Purview: summarize incidents, write queries and reverse-engineer malicious scripts in plain English.
Price Pay-as-you-go: $4 per provisioned SCU/hour
checked 11/15
Google Security OperationsGoogle's SecOps platform (SIEM + SOAR) with Gemini for natural-language search, detection writing and investigation. New (week of Oct 26-Nov 1, 2026): Agentic Chat is in public preview, a rebuilt multi-turn chat for investigating incidents, built on ADK 2.0 with MCP access to SecOps inside the platform. Also in public preview: a Basic mode for SIEM search that shows matching events and field counts without waiting for charts, and the MITRE ATT&CK dashboard now supports ATT&CK v19.1. Preview features fall under Google's Pre-GA terms.
Price See site
See site
CrowdStrike Charlotte AIAgentic security analyst in the CrowdStrike platform that triages detections and speeds up investigations.
Price See site
See site
Splunk AI AssistantNatural-language assistant in Splunk for finding dashboards, reports and data and getting help with searches over your logs.
Price No additional fee for Splunk Cloud Platform and Splunk Enterprise customers
checked 11/15
Mantis (Google)Google's open-source toolkit of skills that lets AI coding agents find, reproduce and patch vulnerabilities in your code, from threat modelling to the final report, plus a reference harness you can point at a repo with a plain-English focus like look for IDOR. About 2,500 GitHub stars. Google says to run it only in isolated environments and to verify every finding by hand.
Price Open source (Apache-2.0); model usage billed by your AI provider
FREEchecked 11/15
Jira Service ManagementAtlassian's IT service desk: request portal, queues, SLAs, knowledge base, incident and change management, with Rovo AI agents and search on paid plans. One of the most common ticketing tools in help desk job posts, and the free plan is enough to build a practice service desk.
Price Free for up to 3 agents; Standard adds Rovo AI agents (per-agent pricing, 14-day free trial)
FREE TIERchecked 11/15
ServiceNow ITSM (Now Assist)ServiceNow's IT service management platform (incident, problem, change and request management) with Now Assist AI that summarizes incidents, suggests fixes and drafts responses, plus AI agents on higher tiers. Shows up in a huge share of help desk and IT job posts. Anyone can get a free Personal Developer Instance with admin access to learn the platform.
Price Quote-based packages (Foundation, Advanced, Prime); free Personal Developer Instance for learning
checked 11/15
GhidraMCPOpen-source MCP server and plugin that lets an AI assistant drive Ghidra, the NSA's free reverse-engineering suite: decompile binaries, list functions, imports and exports, and rename methods and data as it works. About 11,400 GitHub stars; v2.0 shipped Oct 10, 2026 for Ghidra 12.1.4. Analyze malware samples only inside an isolated VM.
Price Open source (Apache-2.0); Ghidra is free; model usage billed by your AI provider
FREEchecked 11/15
StrixOpen-source AI penetration testing agents that run your app, look for vulnerabilities such as injection, broken access control, SSRF and XSS, and confirm each one with a working proof of concept before reporting it, with suggested fixes. Runs locally in Docker with your own LLM API key, and can scan pull requests in CI. About 71,000 GitHub stars; v1.7.0 shipped Oct 5, 2026. Only test apps you own or have written permission to test.
Price Open source (Apache-2.0); Strix Cloud Pro $29/seat/mo, pentests billed separately
FREEchecked 11/15
Shannon (Keygraph)Open-source autonomous AI pentester for web apps and APIs from Keygraph. It reads your source code to find likely attack paths, then runs real exploits against the running app and reports only what it could prove ('no exploit, no report'). Runs from the command line with npx and Docker on your own model provider account. About 49,000 GitHub stars; v3.4.0 shipped Oct 6, 2026. Keygraph says never to point it at production or at anything you aren't authorized to test.
Price Shannon open source free (AGPL-3.0); Keygraph Pro $50/developer/mo
FREE TIERchecked 11/15
PentAGIOpen-source, self-hosted multi-agent system for autonomous penetration testing from VXControl. Describe a target and goal in plain language and a team of AI agents plans the work, runs 20+ tools such as nmap, Metasploit and sqlmap inside an isolated Docker sandbox, remembers what worked and writes a report, with a web UI plus REST and GraphQL APIs. Works with 10+ cloud model providers or local models through Ollama or vLLM. About 32,000 GitHub stars; v2.2.0 shipped Oct 5, 2026. For authorized testing only.
Price Open source (MIT); PentAGI Enterprise available
FREEchecked 11/15
SkillSpector (NVIDIA)NVIDIA's open-source scanner that checks AI agent skills (for Claude Code, Codex, Gemini CLI and MCP) before you install them. It looks for 71 vulnerability patterns across 17 categories, including prompt injection, data exfiltration, privilege escalation, supply-chain risk and MCP tool poisoning, then gives a 0-100 risk score with JSON, Markdown or SARIF reports. Fast static checks work offline; an optional LLM pass adds deeper review. Scans Git repos, URLs, zip files or folders, and runs in Docker. About 24,700 GitHub stars; v2.13.4 shipped Nov 4, 2026 (provider compatibility and source-inspection fixes).
Price Open source (Apache-2.0); optional LLM analysis billed by your AI provider
FREEchecked 11/15
Snyk Agent ScanSnyk's open-source scanner that finds the AI agents, MCP servers and skills already installed on your machine (Claude Code and Desktop, Cursor, VS Code, GitHub Copilot, Windsurf, Gemini CLI, Codex, OpenCode and more) and checks them for prompt injection, tool poisoning, private-data exposure, destructive capabilities, malicious code and hardcoded secrets. Asks before it starts or contacts each MCP server, and Snyk recommends a sandbox or VM for untrusted configs. Runs with uvx or a signed standalone binary and needs a free Snyk API token; component details are sent to Snyk's Agent Scan API for analysis, with secrets redacted. About 3,200 GitHub stars; v0.6.9 shipped Oct 20, 2026.
Price Open source (Apache-2.0); free Snyk account and API token required
FREEchecked 11/15
Cisco Skill ScannerCisco AI Defense's open-source scanner for AI agent skills (Agent Skills format used by Codex and Cursor, plus Claude Code commands with --lenient). It combines YAML and YARA-X rules, code and dataflow analysis and an optional LLM judge that can use cloud or local models, and outputs SARIF for GitHub Code Scanning with a ready-made GitHub Action and a pre-commit hook. Cisco publishes its own measured detection and false-positive rates: rules alone caught 7.7% of malicious test skills at HIGH, and with the LLM judge 66.7% reached MEDIUM or above at a 15.4% false-positive rate, so it says a clean scan is not a guarantee. About 2,600 GitHub stars; 2.3.1 shipped Oct 21, 2026.
Price Open source (Apache-2.0); optional LLM judge billed by your AI provider
FREEchecked 11/15
Strands BoxOpen-source sandbox for AI agents from the Strands Agents project. It runs any agent program inside OS-level isolation and checks its shell commands, Python, network requests and MCP tool calls against default-deny policies (written in the Dogwood policy language), so an agent can only do what a rule permits. Rules can depend on earlier actions, for example blocking an outbound request after the agent read a secrets file. API keys are injected by Box's gateway, so the agent never sees them, and every decision is logged. Written in Rust; preview for Apple silicon Macs, and since v0.2.1 (Nov 5, 2026) also a Linux ARM64 build (glibc 2.34 or newer; the release notes give steps for Amazon Linux 2023). About 500 GitHub stars; v0.1.0 shipped Oct 7, 2026, v0.2.0 and v0.2.1 on Nov 5, with commits daily.
Price Open source (Apache-2.0)
FREEchecked 11/15
1Password for Agents1Password's sign-in layer for personal AI agents, launched Oct 22, 2026. When an agent reaches a login page it asks 1Password for the credential; you see which login it wants and approve or deny on your own device, and 1Password fills the password or one-time code straight into the site, so the agent never sees or stores it. You can disconnect an agent at any time. Integrations for Meta's Muse and Grok Bot are rolling out first; Anthropic's Claude Duet, OpenAI's ChatGPT Work and dots and Town's Townie are listed as coming soon, with Browserbase, Browser Use, Perplexity and others planned. 1Password.com accounts only (not 1Password.eu or .ca), and safeguards vary by integration.
Price No extra cost with 1Password Individual and Families plans
checked 11/15
CVEngineFree, self-hosted external attack-surface scanner with optional AI triage. Add a domain, IP range or URL and it discovers subdomains (bbot), sweeps ports (naabu), fingerprints services (nmap), runs nuclei checks and matches what you run against exploited CVEs from CISA KEV, EPSS, Exploit-DB, Metasploit and other feeds, then confirms what it can with read-only checks that record the exact command they ran. Bring your own AI (Anthropic, OpenAI, OpenRouter, or local Ollama or vLLM) or run it without one; AI-drafted nuclei templates never run until a person approves them. Includes an MCP server and REST API with read-only tokens for agents. One Docker Compose stack on Linux (4GB RAM minimum, 8GB recommended). A fork of reNgine 2.x from a single maintainer, published Oct 24, 2026 and at about 400 GitHub stars two days later. v1.4.0 and v1.4.1 (Nov 14, 2026) make the status page, project rows and headline counts much faster on small servers, draw each registered domain as its own cluster on the graph and skip restarts for documentation-only updates; about 730 GitHub stars by Nov 15. Only scan systems you own or are authorized to test.
Price Open source (GPL-3.0)
FREEchecked 11/15
HOL GuardFree, local-first guard for the AI coding agents you already use. It checks shell commands, file access, package installs and MCP tool calls before they run, flags secret exposure, destructive commands, prompt injection and supply-chain risk, and allows, blocks or asks you to approve each action based on your policy. Works with Claude Code, Codex, Cursor, GitHub Copilot CLI, Gemini CLI, Cline, OpenCode and others through their hooks, so coverage depends on what each agent exposes. Runs without an account, keeps a local history of every decision, can produce an AI bill of materials and SARIF output for CI, and has an optional Guard Cloud for team policies. Install with pipx (Python 3.10+). About 1,000 GitHub stars and roughly 76,000 PyPI downloads in the past month; v3.58.0 shipped Oct 27, 2026.
Price Open source (Apache-2.0); optional Guard Cloud for teams
FREEchecked 11/15
Bumblebee (Perplexity)Free, open-source (Apache-2.0) read-only scanner from Perplexity that answers one incident-response question: when a supply-chain advisory names a bad package, extension or version, which developer machines have it right now? It inventories language packages (npm, pnpm, Yarn, Bun, PyPI, Go modules, RubyGems, Composer, Homebrew), VS Code, Cursor, Windsurf and VSCodium extensions, Chromium-family and Firefox browser extensions, MCP server configs (Cursor, Claude Code, Claude Desktop, Cline, Gemini CLI) and agent-skill lock files, writes NDJSON records and flags exact matches against an exposure catalog. It reads metadata only: it never runs npm, pip or install scripts and does not record the secrets in MCP config env blocks. One static Go binary for macOS and Linux (go install), run once or on a schedule with cron, launchd or MDM; Perplexity maintains exposure catalogs for recent npm supply-chain campaigns in the repo. It is not an EDR, and Codex TOML MCP configs are not parsed yet. About 5,100 GitHub stars; latest tagged release v0.1.2 (Jun 2026), with catalog updates through Oct 2026.
Price Free and open source (Apache-2.0)
FREE TIERchecked 11/15
Vigilo (CERT-EU)Open-source AI penetration-testing and code-review harness built by CERT-EU, the cybersecurity service for the EU institutions, and open-sourced on Nov 13, 2026. Point it at a web application you own: specialised agents, one per vulnerability class (injection, XSS, authentication, SSRF, authorization, GraphQL, WebSocket, cryptography and supply chain), map the target, share findings through a common knowledge base and try to exploit them with captured evidence, then an adversarial critic re-checks every finding, challenges its severity and merges duplicates before it reaches the report. White-box mode reads the source (plus an optional running instance) and proposes a fix branch per confirmed finding; grey-box mode tests only the running app with valid credentials and adds detection hints for the blue team. Runs as a multi-user web platform (Docker Compose, optional GitLab integration) or a command-line pipeline, and drives an agentic coding CLI (Claude Code, OpenAI Codex or OpenCode) with cloud or on-premises models, so you supply the model access. CERT-EU says findings and patches must be validated by the code owners, and it was part of an ENISA proof of concept on AI source-code testing. EUPL-1.2 licence; new on GitHub (only a handful of stars so far). Only test systems you own or are authorized to test.
Price Free and open source (EUPL-1.2); you pay for the model access it uses
FREE TIERchecked 11/15
claude-code-hooksFree, open-source (MIT) collection of ready-made hooks for Claude Code from Karan Bansal, Head of AI at the application security company ArmorCode, shipped as a 22-plugin marketplace you install from inside Claude Code (/plugin marketplace add karanb192/claude-code-hooks, then for example /plugin install block-dangerous-commands@claude-code-hooks). The safety plugins run as separate processes before a tool call executes, so the allow-or-deny decision is made by ordinary code rather than the model: block-dangerous-commands stops commands such as rm -rf ~, fork bombs, curl | sh and force pushes to main; protect-secrets blocks reading, editing or exfiltrating .env and other credential files; git-safety adds branch-aware git guardrails; protect-tests stops the agent deleting or skipping tests to fake a green build; config-guard stops it tampering with its own hook and settings files; instructions-audit locks the session when a loaded CLAUDE.md or rules file carries hidden directives. Other plugins handle session logs, auto-formatting, Slack alerts and context-cost tracking. Every hook is mapped to the OWASP LLM Top 10 2026, and the repo reports 1,949 passing tests. Pattern-based rules can be bypassed (the author says he found two bypasses of his own secrets check) and hooks cannot stop a malicious session-start hook already on disk, so use it alongside scoped credentials and sandboxing, not instead of them. Claude Code only. About 580 GitHub stars; ArmorCode published a write-up on Oct 29, 2026, press coverage followed on Nov 14, and the author plans a Black Hat Europe Arsenal demo on Dec 9-10.
Price Free and open source (MIT)
FREE TIERchecked 11/15
ZendeskWidely used help desk and ticketing platform with an employee service (internal IT and HR) edition. On Oct 26, 2026 Zendesk announced employee service AI agents and advanced action flows that handle onboarding and offboarding, access approvals, laptop lifecycle and badge requests from the help center, Slack or Microsoft Teams, with automations written in plain language. Zendesk says 300 early-access customers ran 6 million workflow executions in five weeks. General availability rolls out Nov 3 to Dec 1, 2026, with 20+ knowledge connectors (Google Drive, SharePoint, Confluence, Notion and more) and automatic escalation to a ticket. From Nov 3, Growth and Professional plans include 10 automated resolutions per AI agent per month, then $0.99 each.
Price Support Team from $19/agent/mo billed annually ($25 monthly); Suite Team $55/agent/mo; 14-day free trial
checked 11/15
BotGaugeSecurity and reliability testing for AI agents you build. It runs adaptive red-team attacks (prompt injection, policy bypass, tool misuse, multi-turn manipulation), replays each failure with a full trace and turns it into a regression test, then keeps watching the agent in production. Connects through an SDK to OpenAI, Anthropic, LangChain, LangGraph, LlamaIndex, CrewAI and AutoGen. Launched as the #3 product of the day on Product Hunt on Oct 28, 2026.
Price See site
See site
Project Quarry (Cisco)Open-source detection rules for AI agent sessions, from Cisco AI Defense. You write rules that match patterns across messages, tool calls, MCP activity and the order they happen in (for example, a read of a sensitive file followed by an outbound send), and a match returns the evidence with secrets masked. Use it as a runtime guardrail, to review past agent traces or during incident response; it ships a plugin for Codex, Claude Code and MCP hosts plus a starter rule pack. Apache-2.0, released Oct 28, 2026, and still alpha, so the rule syntax may change before v1.0.
Price Free and open source (Apache-2.0)
FREE TIERchecked 11/15
Mistral Large 4Mistral's flagship open-weight model (about 1 trillion parameters total, 49B active per token, 1M-token context, text and image input), released as open weights on Oct 29, 2026 and pitched at coding and defensive cybersecurity: finding, reproducing and patching vulnerabilities, malware analysis and writing detection rules. Mistral reports 51% on the Artificial Analysis Cyber Index and a 96.6% average refusal rate on malicious cyber prompts; these are Mistral's own figures. Three reasoning levels (low, medium, high). Far too big for a home PC: self-host on data-center GPUs with vLLM or SGLang, or use it through Mistral Studio, the Mistral API or OpenRouter.
Price API $0.70/M input, $2.10/M output tokens; open weights free (Mistral Ecosystem Open-Weight License; Model-as-a-Service businesses need a commercial license)
FREE TIERchecked 11/15
Radar (Skyhook)Free, open-source Kubernetes UI with a built-in MCP server for AI agents. One binary on your laptop (or a Helm install in-cluster) shows topology, live traffic, an event timeline, Helm and GitOps state, image filesystems and a 44-check cluster audit, with no account. Point Claude Code, Codex, Cursor, Copilot or any MCP client at it and the agent gets trimmed, secret-redacted cluster context instead of raw kubectl YAML; diagnosis is read-only by default and write actions follow your Kubernetes RBAC (or run it with --mcp-read-only). About 4,000 GitHub stars; v1.19.0 shipped Oct 29, 2026, the day it made the Product Hunt top 10. Optional Radar Cloud adds alerts and hosted AI investigations.
Price Open source (Apache-2.0); Radar Cloud free for up to 3 clusters, Team from $149/cluster/mo
FREEchecked 11/15
REA (Reverse Engineer Anything)Free, open-source (MIT) MCP server and CLI that lets your AI coding agent reverse engineer software: native binaries, JavaScript and Electron apps, .NET assemblies and websites, with findings that include the evidence and limits behind each conclusion. Native analysis drives a Ghidra, Hopper, IDA or Kuna install you already have; static JavaScript analysis needs no extra engine, and analysis runs locally. One command (npx rea-agents setup) registers it with Claude Code, Codex, Cursor, Gemini CLI and other agents. The rea-agents npm package was downloaded about 538,000 times in the week of Oct 23-29, 2026 (v6.8.0, Oct 25), and Scientific American profiled it on Oct 30. Only reverse engineer software you own or are allowed to analyze; some users report OpenAI and Anthropic blocking reverse-engineering requests, so check your model provider's terms.
Price Free and open source (MIT); you pay for your own agent or model
FREE TIERchecked 11/15
HackAgentOpen-source (Apache-2.0) red-team toolkit for AI models and agents you own. It runs research-backed jailbreak, prompt-injection and tool-misuse attacks, has judge models score every reply and reports a measured success rate per attack, so you can see whether a fix helped. Use the terminal UI, YAML campaign files you can run in CI, or the hackagent web dashboard. Runs fully local with Ollama and needs no API key; results stay in a local database by default. New in v0.17.0 (Nov 2, 2026): campaigns queued from the dashboard can run on a cloud worker (each run in its own process, with an allowlist of what a campaign may send to the cloud) or locally with hackagent run, plus two new attacks: ToolHijacker tool-document poisoning and execution-verified STAC tool chains. About 535 GitHub stars. For authorized testing only.
Price Free and open source (Apache-2.0)
FREE TIERchecked 11/15
AiSOCFree, open-source (MIT), self-hosted AI Security Operations Center. It ingests telemetry from your security tools (87 connectors including Splunk, Sentinel, Elastic, CrowdStrike, Okta, AWS and Kubernetes audit, plus a syslog/CEF listener), runs detection rules, groups alerts into incidents and has an AI agent triage each one, recording every prompt and tool call in a replayable Investigation Ledger. A human approves any response before it reaches a vendor. It ships a small local model (Llama 3.2 3B via Ollama) so it runs on CPU with no API key, plus an MCP server for Claude, Cursor and Continue and a translator for Splunk SPL, Sentinel KQL and Elastic EQL rules. Needs Docker Compose with 8 GB memory and 20 GB disk. About 7,300 GitHub stars. v22.0.0 (Nov 4, 2026) fixed seven security advisories in its approval and authorization code (two critical). v22.1.0 (Nov 5) fixes more: a critical bug where the documented one-click demo deploy signed session tokens with the default SECRET_KEY published in the repo (anyone who could read the repo could mint any role on those demos), high-severity ways around the connector egress guard (including hostnames that were not shaped like full URLs), password resets that left existing sessions alive and a TOTP code that could open more than one session, plus medium issues around committed .env backups in git history and agent tool arguments that were not checked against each tool's schema. Update if you run it, and set your own SECRET_KEY on any deployed instance. v22.2.0 (Nov 5) is a reliability release: a fusion worker that lost its message broker now retries on its own instead of staying down until someone restarted the container, and a new CI check makes every connector read its full poll window (six connectors had shipped a paging bug that could drop alerts). The project publishes its own audit of what does and doesn't work yet.
Price Free and open source (MIT)
FREE TIERchecked 11/15
RedAmonFree, open-source (MIT), self-hosted AI penetration testing framework. It runs recon tools in parallel and maps your attack surface into a Neo4j graph, then an AI agent tries to exploit what it found from a Kali sandbox, with human approval gates before risky steps, and a CodeFix agent can open GitHub pull requests that fix the issues. Works with MCP in both directions (plug in MCP servers as tools, or drive RedAmon from Claude Code). Runs in Docker; needs about 7 GB RAM and 80 GB disk without OpenVAS. The project reports solving 101 of 104 XBOW web-security benchmark challenges black-box; that is its own claim. About 3,300 GitHub stars. For systems you own or have written permission to test only.
Price Free and open source (MIT)
FREE TIERchecked 11/15
PyRIT (Microsoft)Microsoft's open-source AI red teaming toolkit (the Python Risk Identification Tool) for testing generative AI apps and models: automated scans, adaptive multi-turn attacks such as jailbreak and prompt-injection attempts, scoring of what got through, and CoPyRIT, a web UI. Built by Microsoft's AI Red Team. About 4,700 GitHub stars and active daily; v1.0 shipped Jul 2026 and v1.1.0 on Sep 4, 2026. Only test AI systems you own or are authorized to test.
Price Open source (MIT); model usage billed by your AI provider
FREEchecked 11/15
Deckard (Arthur)Free, open-source (MIT) scanner from Arthur that finds the AI agents running across your organization (shadow AI) and sends each one to your SIEM. It checks endpoints with osquery queries from Arthur's open ai-detection-spec catalog (Claude Code, Claude Desktop, Cursor and more), cloud AI platforms such as Vertex AI Agent Engine and AWS Bedrock and AgentCore, cloud network logs and your Splunk or Elastic data, then writes one event per agent found to Elasticsearch, Splunk HEC or any webhook. Headless and read-only, and it runs entirely in your environment; the quickstart scans your own machine with Docker Compose and osquery and shows the results in a local Kibana, no account needed. Launched Nov 3, 2026 (v0.2.1 Oct 27); about 170 GitHub stars on launch day, with commits daily.
Price Free and open source (MIT)
FREE TIERchecked 11/15
Failproof AILocal guardrails and session history for the AI agents you already run. It hooks 12 agent harnesses (Claude Code, Codex, GitHub Copilot CLI, Cursor Agent, OpenCode, Factory Droid, Devin CLI, Antigravity CLI, Goose, Pi, Hermes and OpenClaw), records every run and blocks dangerous tool calls before they execute using 40 built-in policies or rules you write in TypeScript. A local dashboard on localhost:8020 needs no account, and failproofai audit scans your history for risky patterns and suggests policies. Blocking is verified on all 12 harnesses; what else it can stop varies by harness. Install with npm. About 7,600 GitHub stars and about 2,300 npm downloads in the week of Oct 27-Nov 2, 2026; v1.0.13 (Nov 2) fixed policies that had silently stopped applying on newer OpenClaw and Antigravity versions. A hosted Observability product for teams is sold by demo.
Price Free for personal and internal use (MIT with Commons Clause); hosted Observability for teams by demo
FREE TIERchecked 11/15
AI-Infra-Guard (Tencent)Free, open-source (Apache-2.0) AI red teaming platform from Tencent's Zhuque Lab, also called A.I.G. Five scanners in one web UI: AI infrastructure scanning that fingerprints AI services on a domain, IP or URL and matches known vulnerabilities, MCP server scanning, agent skill auditing, agent scanning that sends test prompts to a running agent, and LLM jailbreak evaluation of any OpenAI-compatible model. Start it with Docker Compose (4 GB RAM, 10 GB disk) and open localhost:8088, or pip install each scanner on its own (aig-infra-scan, aig-mcp-scan and others). The MCP, skill and agent scans use an LLM API key you supply. It has no login, so Tencent says never to expose it on a public network. About 7,500 GitHub stars, weekly releases (v4.6.9 on Nov 5, 2026 adds a CloudGPU provider preset for agent scans and MCP-scan fixes) and a Black Hat Europe 2025 Arsenal demo. v4.6.10 (Nov 12, 2026) adds a local generative AI safety self-check, one page that runs refusal checks and infrastructure scans and keeps a printable report without storing your API keys, plus about 135 new AI product CVE advisories from NVD and several MCP-scan and agent-scan fixes. Only scan systems you own or are authorized to test.
Price Free and open source (Apache-2.0); LLM usage billed by your AI provider
FREE TIERchecked 11/15
CISO AssistantOpen-source GRC (governance, risk and compliance) platform from intuitem for risk assessments, audits, vendor risk and compliance tracking against 200+ frameworks, including ISO 27001, NIST CSF, SOC 2, CIS Controls, PCI DSS, HIPAA and CMMC, with automatic mapping between frameworks so one control counts toward several. An optional AI chat mode (off by default) answers questions about your risks, controls and frameworks using a local model through Ollama, LM Studio or llama.cpp, or any OpenAI-compatible endpoint, with a Qdrant vector store. Self-host the free Community edition with Docker Compose, or start a free trial of the cloud version. About 4,600 GitHub stars. v4.2.2 (Nov 7, 2026) adds MCP tools that let an AI assistant assess and write findings, an assets lifecycle, a cumulative outcome for Australia's Essential Eight, new framework libraries (ASD ISM September 2026, ASD Essential Eight October 2026, Swift CSCF 2026 and 2027, SAMA BCM, Germany's 2025 NIS-2 implementation act), a better ISO 27001 to NIST CSF 2.0 mapping and security hardening. v4.3.0 (Nov 13, 2026) adds service accounts federated with an external OIDC identity provider, an MCP tool for reviewing evidence, a more actionable landing-page briefing, a Hungarian NIS2 compliance preset and requirement names for the NIS2 2024/2690 framework, and tightens bulk-action permissions. v4.3.1 (Nov 14, 2026) makes the AI chat index evidence attachments and published documents, fixes MCP sign-in failing when an app requests scopes it may not have, adds an outline of ISO 9001:2015 for quality management (English and French) and improves how tasks are shown.
Price Community edition free (self-hosted, AGPLv3); Pro SaaS from €39/contributor/mo billed annually; free cloud trial
FREE TIERchecked 11/15
Codex Security (OpenAI)OpenAI's security scanner for application code: find, validate and fix vulnerabilities with a Codex plugin, a terminal CLI (@openai/codex-security) and a TypeScript SDK, plus a research-preview Codex Security Cloud for connected GitHub repos. Scan a repo, selected paths or a git diff; deep scans repeat standard passes and dedupe findings; import Semgrep, Snyk or Wiz results; export SARIF, JSON or CSV. Runs with an OpenAI API key (or Codex API key) and can also use other providers such as Amazon Bedrock. About 11,300 GitHub stars (Apache-2.0); npm package at 0.8.3 as of Nov 3, 2026. Some cybersecurity requests need Trusted Access for Cyber. Only scan code you own or are authorized to review.
Price Open source (Apache-2.0); OpenAI inference needs Codex Security / API access (usage billed by your OpenAI account); other providers supported
FREEchecked 11/15
CVE IntelFree, open-source desktop app for vulnerability triage, written in native Rust, that is also a CLI, REST API and MCP server (41 tools) in one executable. Give it a CVE ID, a project folder, lockfile or SBOM, a container image, the software on your own computer, your own domains, or a suspicious IP, hash or domain; it queries up to 41 public sources (NVD, CISA KEV, FIRST EPSS, ENISA EUVD, OSV, GitHub Advisories and more, many with no API key) and returns a 0-100 risk score, whether attackers are already exploiting it, the evidence behind each signal and what to do next. Reads and writes CycloneDX SBOMs and keeps a decision ledger. An optional Ask AI assistant runs on a small local model (Qwen 2.5 3B via llama.cpp by default, about 6 GB RAM) or Ollama/LM Studio; remote providers are opt-in. Installers for Windows, macOS and Linux (not code-signed yet, so expect a first-launch warning). About 137 GitHub stars. v0.3.27 shipped Nov 2, 2026 (Arch Linux package scanning). v0.3.28 (Nov 6, 2026 ET / Nov 7 UTC) adds OAuth 2.1 and HTTPS for `cve-intel serve` and refuses to start on a non-loopback address (including Docker's 0.0.0.0) unless OAuth is configured or you set CVE_INTEL_ALLOW_UNAUTHENTICATED=1; the desktop app and serve on 127.0.0.1 are unchanged.
Price Free and open source (MIT or Apache-2.0)
FREE TIERchecked 11/15
Grim PrompterFree, open-source (MIT) threat-modeling skill for AI coding agents from Jim Manico, the application security trainer and longtime OWASP contributor. Link its threat-model skill into Claude Code or Codex (or add a snippet to AGENTS.md for any other agent), then ask for a quick model, a full six-stage deep analysis, one method such as STRIDE per element, or a review of how a branch changes an existing model. 36 architecture workflows and 29 method and analysis prompts; the runbook requires read-only inspection with file-and-line citations, and builds, tests, installs or network access need your explicit approval. Results land in a threat-model/ folder as a Markdown report plus JSON you can diff later. Also covers MCP servers and agent tools (credentials, permissions, tool-description poisoning). About 400 GitHub stars. v1.0.0, the first stable release, shipped Nov 8, 2026, with engineering-first privacy workflows, optional LINDDUN methods and GDPR and CCPA/CPRA profiles. Output is a draft for you to review, not a security certification.
Price Free and open source (MIT); you pay for your own coding agent
FREE TIERchecked 11/15
StroqFree, open-source (Apache-2.0) local security CLI for AI coding agents. Run npx @stroq/cli and it reads your last 30 days of Claude Code and Codex sessions and shows how many commands came from text the agent read and which keys left your machine; stroq sent checks Claude Code, Codex CLI and Cursor sessions against the credential files on your computer and reports where each key went, printing key names, never values. stroq init installs hooks (Claude Code, Cursor, Codex, Copilot CLI, Windsurf, Antigravity, plus an MCP proxy) that deny curl-pipe-to-shell and outbound calls carrying a known secret, and ask you first when a command repeats text the agent read. A local console on 127.0.0.1 shows every decision with its reason. No account, no telemetry; needs Node.js 22+. Matching is text-level, so it is a filter, not a sandbox. About 225 GitHub stars and about 600 npm downloads in the week of Nov 1-7, 2026; v0.30.0 shipped Nov 5 alongside a Show HN post. v0.31.0 (Nov 8) adds stroq inspect, which reads a folder someone sent you for code that runs when it is fetched or decoded, code hidden far to the right of a line, scripts that fetch into a shell and config files a tool runs by itself (places to read, not verdicts), and the hook now asks the first time an agent runs code in such a folder; v0.31.1 (Nov 9) fixes stroq sent missing an environment dump inside a Python f-string URL. v0.32.0 (Nov 11) reads Cursor sessions on its card alongside Claude Code and Codex (read-only), and lets you mark a flagged moment as a false alarm from a terminal with stroq show --false-alarm; the mark is dropped again if the evidence changes, and the hook refuses marks made through an agent's shell.
Price Free and open source (Apache-2.0)
FREE TIERchecked 11/15
sandbox-mcpFree, open-source (MIT) tool from Scott Hanselman that gives you or an AI agent a fresh, disposable Windows 11 machine built on Windows Sandbox (needs Windows Pro or Enterprise): boot it, deploy an app or installer, drive the desktop with UI Automation, capture screenshots, recordings, crash events and exit codes, then throw the machine away. The sbx command-line tool works on its own and an optional MCP server exposes the same engine to agent clients. Guest networking can be off, firewalled or on, which makes it useful for the classic works-on-my-machine check on a clean PC and for looking at an untrusted installer without touching your real system. It runs one machine at a time and does not promise every Windows app will run in Sandbox. About 355 GitHub stars; v1.0.0 shipped Nov 6, 2026 and v1.2.0 on Nov 7. Very new.
Price Free and open source (MIT); needs Windows Pro or Enterprise
FREE TIERchecked 11/15
0 (0sec Labs)Free, open-source AI security harness from 0sec Labs (a Swiss lab backed by Y Combinator) for code review, pentesting, vulnerability research, reverse engineering, verifying findings and drafting fixes. Run 0 and it opens a local web console (no account needed) with chat, a findings queue and board, workflow templates for assessment, verification, fixes and research, reusable skills and source-linked memory; the same workflows run from a headless CLI (0 workflow run, with SARIF and other reports) or as an MCP server so Claude Code, Codex and other agents can call them. It is not a model: you connect your own model provider or a supported subscription. Binary analysis uses Ghidra through its 0verse add-on. Installs on Apple silicon Macs and x64/ARM64 Linux (npm, Windows, Docker and source installs are documented). Its README says the default shell runs on your host, so use an isolated executor for untrusted work, and only assess systems you are authorized to test; an optional scope plugin enforces target boundaries. About 1,190 GitHub stars; the repo dates from August 2026 but its public open-source launch was Nov 8, 2026, and it ships several releases a day (v0.28.9 on Nov 9). Very new, so treat it as early software.
Price Free and open source (MIT or Apache-2.0); you pay your model provider. Hosted 0cloud PR reviews: $100 free trial credits, then $1/credit ($5 per PR security test)
FREE TIERchecked 11/15
Neoverse (Neo)Free searchable catalog from Neo Security that gives AI add-ons a security verdict before you install them: MCP servers, agent skills, desktop apps, and browser and IDE extensions (VS Code, JetBrains, Cursor, Windsurf, the Chrome, Edge and Firefox stores, GitHub, the MCP registry and package managers such as Homebrew, Winget and Chocolatey). Neo says it has classified more than 4 million items; each analyzed one is marked Malicious, Unsafe or Safe with the headline finding behind the verdict, a summary and version history, based on static analysis, a sandboxed run, a check of the app's behavior against its own docs, plus CVE and GitHub advisory data, repeated for each new version. Free with email sign-in; full findings, evidence and API access are for Neo customers or on request. Neo says signing up does not protect your machine: it is a lookup, not a blocker. Public launch Nov 9, 2026. Verdicts come from Neo's own automated research, so treat them as one signal and still read what a tool can access.
Price Free catalog with email sign-in; full findings and API access for Neo platform customers
FREE TIERchecked 11/15
Elastic Security AIAI in Elastic Security: automated alert triage and hypothesis-driven threat hunts with ES|QL over your SIEM data.
Price Elastic AI Assistant included in Serverless Security Analytics Complete, from $0.11/GB ingested plus retention
checked 11/15