Superagent is an open-source safety layer for AI applications. It protects LLM apps against prompt injections, data leaks and harmful outputs, and helps teams prove compliance to their customers โ with safety embedded directly in the request path rather than bolted on afterwards. The project has about 6,800 stars on GitHub and was updated in 2026.
Every input to and output from your AI app passes through Superagent’s guardrail models and classifiers: user prompts are screened for injection and jailbreak attempts before they reach your model, and responses are checked for leaked secrets, personal data and policy-violating content before they reach the user. Redaction can mask sensitive data instead of blocking outright, and detections are logged so security teams get an audit trail of attacks and near-misses. Superagent is delivered as an API and SDKs that sit in front of providers such as OpenAI and Anthropic, and the core is open source under the MIT license, so it can be self-hosted or run as their managed service. The company (formerly known for an open-source agent framework of the same name) now focuses entirely on this AI security product, targeting teams that must ship LLM features under real compliance requirements โ SOC 2 conversations, enterprise security reviews and regulated-industry rules.
Superagent’s core is free and open source under the MIT license for self-hosting. The managed cloud service offers a free tier with paid plans based on usage for teams that want hosted guardrails and support.
Superagent is for teams shipping LLM features to real users โ especially in enterprises and regulated industries โ who need injection defense, leak prevention and an audit trail without building safety infrastructure themselves.