Teleport issues short-lived certificates for SSH, Kubernetes, databases, and cloud consoles instead of handing out static credentials, with session recording and just-in-time privilege elevation built in rather than bolted on. Install it with the goteleport.com script or a Docker image, then route infrastructure access through the Teleport Proxy. It's the reference implementation for infrastructure PAM at this point, with nearly 28,000 commits behind it.
Pricing: free and open source (AGPL-3.0) if you build from source; compiled binaries and container images moved to a commercial license starting with v16.
Vault centralizes secrets management — dynamic credentials, encryption as a service, PKI issuance, and an audit trail of who touched what. docker run hashicorp/vault server -dev gets a sandbox running in seconds; production deployments pull a binary from releases.hashicorp.com. It's the tool OpenBao forked from once Vault moved off a fully open license, and most teams still reach for it first.
Pricing: source-available under the Business Source License 1.1, converting to MPL-2.0 four years after each release — free to run, but you can't resell it as a competing hosted service.
Cerbos pulls authorization out of application code and into YAML policies that a sidecar or standalone service evaluates per request, which matters as much for AI agents checking tool permissions as for the web apps this pattern was built for. docker run cerbos/cerbos:latest gets a decision point running immediately, with SDKs for eight languages and a Helm chart for production. v0.57.0 shipped October 23, 2026, and releases land roughly weekly.
Pricing: the policy decision point is Apache 2.0; Cerbos Hub adds paid management and audit tooling on top.
Pocket ID is a self-hosted OIDC provider that drops passwords entirely — users sign in with a passkey, full stop. A docker-compose.yml from the repo gets an instance running against your own apps within minutes. It trades breadth for a small footprint next to something like Keycloak, and 9,800+ stars suggest plenty of people are fine with that trade. v2.19.1 landed October 22, 2026.
Pricing: free and open source, BSD-2-Clause.
Caido is a Rust-backed intercepting proxy for web app testing — capture traffic, replay and tamper with requests, chain repetitive steps into automation workflows. It ships as a desktop app for Windows, macOS, and Linux, and now comes bundled in Kali. The free Basic tier caps you at two projects and one active session; paid tiers unlock unlimited projects and an AI assistant.
Pricing: freemium — free Basic tier, paid Individual/Team/Enterprise tiers, a free one-year plan for students and teachers.
AdaptixC2 is a modular command-and-control framework for red-team engagements, pairing a Go server with a cross-platform Qt client, pluggable listeners and agents, and a scripting engine for automating post-exploitation steps. Build it with the repo's install script or docker-compose; the README states plainly that it's for authorized testing only. It's newer than Sliver or Mythic but shipping fast, past 780 commits already.
Pricing: free and open source, GPL-3.0.
Bishop Fox built CloudFox to answer "what can I actually reach from here?" during a cloud pentest — it enumerates AWS, Azure, and GCP resources and surfaces privilege-escalation paths, exposed secrets, and network misconfigurations, roughly PowerView for the cloud. go install github.com/BishopFox/cloudfox@latest, or grab a binary from Releases. It's the established pick in this niche and still gets patched regularly.
Pricing: free and open source, MIT.
Hayabusa turns Windows Event Logs into a readable hunting timeline fast, running 4,000+ community Sigma rules alongside roughly 170 of its own detections against EVTX files. Grab a signed binary from Releases and point it at a log export — no server required. The tool itself is AGPLv3, but the bundled rule set ships under a separate Detection Rule License that's worth reading before redistribution.
Pricing: free and open source (AGPL-3.0 for the tool; rules under DRL 1.1).
Timesketch is Google's collaborative timeline platform for incident response — upload a Plaso supertimeline or a CSV, then search, annotate, and let built-in analyzers flag lateral movement or credential access across a shared case. The install guide walks through a Docker setup, and multiple analysts can work the same timeline simultaneously. It tagged release 20261014 this month.
Pricing: free and open source, Apache-2.0.
Beelzebub swaps canned honeypot responses for an LLM that improvises plausible system output across SSH, HTTP, TCP, and Telnet traps — plus an MCP honeypot built to catch prompt-injection attempts aimed at agents rather than humans. YAML defines each trap; ./install.sh --docker gets it running. The MCP honeypot is new this year, which tells you where deception tooling is heading as more attack surface shifts into agent tool calls.
Pricing: free and open source, GPL-3.0.
BBOT chains subdomain enumeration, port and web scanning, cloud-bucket discovery, and breach-data checks into one recursive recon pipeline, so a scan builds on what the previous module found instead of running every step in isolation. pipx install bbot, then bbot -t yourdomain.com -p subdomain-enum starts a run. Black Lantern Security has kept it shipping since 2022, with v3.0.2 out this week.
Pricing: free and open source, AGPL-3.0 — a copyleft license worth checking before embedding it in a hosted service.
Sherloq is a Qt-based image forensics bench for the part of OSINT that isn't domain lookups — tampering detection, compression-artifact analysis, PRNU fingerprinting, and EXIF anomaly checks on a single photo. Clone the repo, install the Python requirements, run sherloq.py for the GUI. It's been around since 2017; its latest commits bumped the bundled ExifTool and fixed PRNU calculations.
Pricing: free and open source, GPL-3.0.
Consensus-OSINT folds 19 separate username, email, and phone lookup projects into one CLI with bulk search and evidence graphs, so one command replaces stitching a dozen tools together by hand. pip install "consensus-osint[modules,pdf]", then consensus email "address@example.com" is the entire setup. It's a few weeks old as of this writing, maintained by a community of OSINT practitioners rather than one person.
Pricing: free and open source, GPL-3.0-or-later.
Session routes messages through its own onion network so no central server ever learns who's talking to whom, and signup needs nothing beyond a generated account ID — no phone number. Install the desktop or mobile app, or build from source to audit it yourself. Session Protocol v2, announced October 13, 2026, adds post-quantum crypto on top of existing forward secrecy.
Pricing: free and open source, GPL-3.0.
AWS Labs built ferret-scan to catch PII, API keys, and other secrets in files before they leak, scoring each match by context and redacting it while preserving the surrounding format. It runs entirely offline — nothing scanned leaves the machine — as a single Go binary, a pip install, or a pre-commit hook that outputs SARIF for CI. It's a recent addition to the AWS open-source labs catalog.
Pricing: free and open source, Apache-2.0.
OpenMLS implements the IETF's Messaging Layer Security protocol (RFC 9420) in Rust — the newer standard for scalable group encryption several messaging vendors are building toward instead of stretching Signal's double-ratchet design across large group chats. Add it as a Cargo dependency and swap in whichever crypto provider you want. Version 0.9.1 shipped October 7, 2026.
Pricing: free and open source, MIT.
Scorecard checks a GitHub-hosted repo against roughly twenty supply-chain hygiene signals — pinned CI dependencies, branch protection, signed releases, fuzzing, dangerous workflow patterns — and returns a 0-to-10 score plus the detail behind it. Run scorecard --repo=github.com/org/repo, install via Homebrew, or drop the official GitHub Action into a workflow to score every PR. It's become the default input other dependency-risk tools build policy on top of.
Pricing: free and open source, Apache-2.0.
Witness wraps each build-pipeline step to produce a signed in-toto attestation of what actually ran, then checks an embedded OPA policy to confirm an artifact came from an authorized sequence of steps before release — a layer below signing, closer to provenance than to the signature itself. One install script and witness run -- <your build command> gets attestations flowing. It's a CNCF in-toto project, donated by TestifySec.
Pricing: free and open source, Apache-2.0.
Harden-Runner watches GitHub Actions runners at the eBPF level for unexpected network egress, file writes, and process activity — it's the tool that caught the Trivy build compromise as it happened. Add uses: step-security/harden-runner@v2.22.3 with egress-policy: audit as the first step in a job. The free tier covers public repos on GitHub-hosted runners; private repos and self-hosted runners need the paid tier.
Pricing: freemium — free for public repos on GitHub-hosted runners, paid Enterprise tier for private repos and self-hosted/Kubernetes runners.
Strix runs autonomous agents that recon an application, attempt exploitation, and validate findings with a working proof instead of a flagged line in a report — a SAST/DAST hybrid built around AI agents rather than signature rules. curl -sSL https://strix.ai/install | bash, then point it at your own codebase; it runs locally, in CI, or air-gapped. It launched in mid-2025 and has picked up over 72,000 stars since.
Pricing: free and open source, Apache-2.0.
Cloud Custodian runs YAML rules that query, filter, and — if you choose — automatically remediate misconfigured resources across AWS, Azure, GCP, and Kubernetes: untagged instances, unencrypted buckets, public security groups. pip install c7n, then custodian run policy.yml fires a policy against a live account. It's a CNCF Incubating project that's been doing this since 2015, with v0.9.52.0 out in September.
Pricing: free and open source, Apache-2.0.
Gatekeeper is a Kubernetes admission controller built directly on Open Policy Agent — define a ConstraintTemplate and Constraint as CRDs, and it validates or mutates incoming resources against that policy, with an audit mode that flags drift in what's already running. helm install gatekeeper/gatekeeper -n gatekeeper-system is the whole deployment. It's the OPA-native admission-control project, CNCF Graduated, with a v4 GA in the works.
Pricing: free and open source, Apache-2.0.
KubeArmor enforces least-privilege behavior on containers and nodes through Linux Security Modules — BPF-LSM, AppArmor, SELinux — rather than just watching kernel events and alerting after the fact, which is what separates it from detection-only eBPF tools. karmor install, or the Helm chart from kubearmor/charts, gets policies enforcing. It's a CNCF Sandbox project on its 1.7.3 release.
Pricing: free and open source, Apache-2.0.
ToolHive runs MCP servers inside isolated containers with per-request access policy, network filtering, and secrets injection, plus a gateway adding OIDC/OAuth SSO and OpenTelemetry metrics for anyone running MCP servers in production instead of on a laptop. Install the CLI or desktop app from stacklok.com, or deploy the Kubernetes operator for a fleet of servers. Stacklok has pushed to it daily since early 2025.
Pricing: free and open source, Apache-2.0.
Microsoft's toolkit sits in front of autonomous agents as a runtime policy layer — sub-millisecond enforcement, execution sandboxing, and a compliance score mapped against all ten risks in the OWASP Agentic AI Top 10. pip install "agent-governance-toolkit[full]" covers Python; Rust, Go, .NET, and TypeScript bindings also ship. Microsoft open-sourced it in April 2026, and it's already past 6,500 stars.
Pricing: free and open source, MIT.
Agent OPFOR red-teams AI agents and MCP servers specifically, rather than probing a single LLM call in isolation — it generates prompt injections, jailbreaks, tool-misuse attempts, and multi-turn reasoning attacks against your agent, then has a separate LLM judge score what got through. It runs from the CLI, an IDE plugin, or a browser extension. The project is six months old and already past 500 stars.
Pricing: free and open source, Apache-2.0.
Hack The Box is the practice range most people mean when they say they're "doing a box" — over 1,500 retired and active machines, live CTF events, and Pro Labs that simulate a full corporate network rather than one isolated target. Sign up free at hackthebox.com/get-started; no card is needed for starting machines and a chunk of Academy content. An AI Range for benchmarking agentic attackers launched October 6, 2026.
Pricing: free tier plus paid VIP+ (around $25/month) and the newer PRO tier (around $49/month) adding ten flagship Pro Labs.
GOAD stands up a deliberately broken multi-domain, multi-forest Active Directory lab — Kerberoasting, AS-REP roasting, ACL abuse, a stack of ADCS escalation paths — and it's the lab most AD-focused courses point students toward by now. Clone the repo and run ./goad.sh, which drives Vagrant and Ansible (or Docker/Terraform, if you'd rather) to build the whole domain on your own machine. Orange Cyberdefense has kept it current, with commits as recent as five days ago.
Pricing: free and open source, GPL-3.0.
Hackversity spins up an isolated, auto-terminating lab for each challenge in about a minute, across eight categories from web exploitation to reverse engineering, instead of making you set up a local VM for every exercise. Sign up with email or Google — no card needed for the free tier's community challenges and learning paths. It launched on Product Hunt on October 14, 2026, so the content library is still thinner than the established platforms.
Pricing: free community tier; paid Unlimited tier removes the lab-session cap and unlocks Hard/Insane challenges, plus pay-per-use credits for hints.