Installing an open-source package usually means trusting that nobody checked whether it or one of its dependencies has a known security hole, and that risk has gotten worse now that AI coding agents can install packages on their own without a human double-checking each one first. Refuse works like a security guard standing at the door of your package manager: before npm install, pip install, or any of 16 other install commands actually run, it checks the exact package and version against a huge, current database of known vulnerabilities, and simply refuses the install — telling you which safe version to use instead — if there's a match. Refuse is a free, open-source (Apache-2.0) pre-install shim wrapping 18 package managers (npm, pnpm, yarn, bun, npx, pip, pip3, uv, poetry, pipenv, pdm, pipx, cargo, gem, bundle, go, composer, dotnet). It intercepts install commands and checks the target package/version against 362,000+ security advisories across 21 ecosystems, blocking installs that match a known vulnerability and suggesting a safe alternative version. It was built specifically in response to a documented surge in 2026 supply-chain attacks targeting AI coding agents that auto-install dependencies without review. It's self-hostable as a single Docker container with your own CVE backend, and collects no telemetry.
| # | Tool | Best for | Price |
|---|---|---|---|
| 1 | Entreprises SaaS B2B qui doivent proposer le SSO entreprise à leurs clients grands comptes sans refaire leur système d'authentification | — | |
| 2 | Équipes sécurité et plateforme qui déploient des agents IA autonomes multi-fournisseurs, entreprises soumises à conformité (SOC 2, HIPAA, EU AI Act) | — | |
| 3 | Organisations déployant des agents IA et serveurs MCP connectés à des systèmes d'entreprise | — | |
| 4 | Particuliers et petites équipes qui veulent une protection anti-phishing supplémentaire dans le navigateur, sans configuration | — | |
| 5 | Équipes sécurité d'entreprise qui veulent un pentest continu et prouvé (pas juste des rapports annuels), y compris sur les risques spécifiques aux applications IA | — | |
| 6 | Équipes dev et sécurité qui veulent scanner leur code et leurs dépôts pour des secrets exposés (clés API, mots de passe) dans leur pipeline CI/CD | — | |
| 7 | Journalistes, militants et utilisateurs ayant des besoins de confidentialité élevés qui veulent éliminer toute trace de métadonnées, pas seulement chiffrer le contenu des messages | — | |
| 8 | Équipes de développement qui veulent une couverture de pentest continue en CI/CD sans le coût d'un audit manuel régulier | — | |
| 9 | Particuliers voulant des conversations IA privées, développeurs, organisations manipulant des données sensibles (santé, légal, finance) | — | |
| 10 | Startups et PME tech qui doivent obtenir SOC 2, ISO 27001, HIPAA ou GDPR pour signer des clients entreprise, et veulent éviter la collecte de preuves manuelle | — | |
| 11 | Équipes sécurité/DevSecOps voulant des tests d'intrusion continus et automatisés en complément (pas remplacement) d'audits humains | — | |
| 12 | Utilisateurs mobiles soucieux de leur vie privée, y compris personnes à risque de surveillance ciblée (journalistes, militants) | — |
Big customers often refuse to sign a contract unless your app supports "enterprise SSO" — logging in through their own Okta or Azure AD instead of a normal password. Building that yourself for every possible identity provider can take months; SSOJet plugs
When you let an AI agent act on its own — call APIs, touch databases, send emails — you lose the ability to watch over its shoulder the way you would a human employee. Lineation sits between your agents and the systems they touch, checking every action ag
When you connect an AI agent to your company's tools through MCP (the protocol that lets Claude, Cursor and similar assistants call real APIs), you're opening a new door into your systems — and most teams have no way to watch what walks through it. Gopher
A free browser extension that quietly checks whether a link or website is actually what it claims to be, before you click, without asking you to configure anything.
An AI agent that behaves like a real hacker against your own applications — chaining exploits together automatically, proving they work, and telling your team exactly what to fix, continuously instead of once a year.
A free, open-source scanner that catches API keys and passwords accidentally committed to your code — built by the same developer who created the widely-used Gitleaks, after he lost control of that project's name.
A free messaging app, like WhatsApp or Signal, but designed so it doesn't even know who you are — there's no phone number, username or account ID tied to your identity, just private connections you make with people directly.
A free tool that automatically tries to hack into your own web application to find security holes before a real attacker does, then proves each vulnerability actually works instead of just guessing at it.
Every AI provider asks you to trust that they won't peek at your data — Tinfoil instead runs your AI chats and API calls inside a locked hardware box (a "secure enclave") that's mathematically provable to be sealed off, even from Tinfoil's own engineers,
A service that automates most of the tedious work of getting and keeping security certifications like SOC 2 — connecting to your tools, collecting proof continuously, and prepping you for the audit — instead of chasing screenshots in spreadsheets.
An open-source platform that runs automated penetration tests using a team of 18 AI agents coordinating 80+ security tools — probing your web apps, cloud, and infrastructure for real, exploitable vulnerabilities the way a human pentest team would, but con
