Superagent

Superagent has evolved from its earlier agent-workflow positioning into a security platform for AI-native applications and coding agents. The current product surface covers pull-request scans, contributor trust, runtime guardrails, context guardrails and authorized red-team testing, with access through GitHub, REST API, MCP, CLI and supported coding-agent workflows.¹²

Runtime Guardrails are designed to monitor or enforce policies around agent activity, including suspicious commands, secrets and sensitive data. Superagent states that the normalized event stream remains on the endpoint by default while findings, enforcement decisions and client health can be sent to the service depending on configuration. The initial documented endpoint support is macOS and Linux rather than Windows.³

Context Guardrails scan material before an agent consumes it, including web pages, email, files, agent skills, MCP repositories and packages. Red Team provides black-box and gray-box adversarial testing against repositories, web applications, agents and packages, with reproducible findings and isolated execution; the service terms restrict this testing to systems the customer owns or is authorized to assess.⁴⁵⁶

The public Superagent SDK is MIT-licensed and exposes guard, redact, scan and red-team functionality through TypeScript, Python, CLI and MCP-oriented workflows. Hosted services and the broader platform remain proprietary, so the product is partly rather than wholly open source.⁷

Superagent publishes unusually specific pricing for a security platform. The private free tier includes 100 credits per month. Starter is $499 per month for 500 credits, Growth is $1,499 for 2,000 credits and Scale starts at $3,999 for 6,000 credits. Public and open-source repositories are free with unlimited usage, while annual paid plans include 20% more credits.⁸

All current plans include the product suite, API/MCP access and unlimited users. Credit consumption varies by function: the published table lists one credit for a PR Scan or Context Guardrails item and 500 credits for a Red Team report, while some runtime and contributor-trust functionality is included. That makes the free and open-source access attractive for evaluation and public projects, but sustained adversarial testing can consume paid allocations quickly.⁸

The strongest evidence for Superagent is its current documentation, public code and legal disclosures rather than independent performance testing. Y Combinator independently identifies Superagent as an active Winter 2024 company focused on security for AI-native developers, which supports company identity and current positioning but is not evidence of detection accuracy or reliability.⁹

The Services Agreement is explicit about important limits: automated outputs may be incomplete or inaccurate, the service does not guarantee security or compliance, and customers must maintain human oversight. It also prohibits unauthorized security testing and restricts intentional submission of regulated sensitive data such as PHI, PCI data, biometric data, government identifiers and children's data unless separately agreed with appropriate safeguards.⁶

For customer data, the agreement states that Superagent will not use it to train, fine-tune or improve a generalized machine-learning model unless the customer expressly authorizes that use. The Privacy Policy and global DPA provide the current public privacy and processor framework.⁶¹⁰¹¹

This is a safety-sensitive product because it can inspect secrets, personal data, code and agent actions and can perform adversarial testing. Endpoint-local runtime processing, explicit authorization boundaries and enforcement controls are positive design signals. However, buyers should still validate retention, subprocessors, access controls, incident response and the evidence behind security claims for their deployment. No current exact-product independent benchmark or substantial user-review corpus was found that could validate Superagent's efficacy numerically; missing external evidence is not treated as evidence of poor performance.

Superagent is best suited to developers, platform teams and security teams deploying coding agents or AI-native applications and wanting controls at several points in the lifecycle: code review, pre-consumption context scanning, endpoint/runtime enforcement and sanctioned security testing. Its GitHub, API, MCP and coding-agent integrations make it especially relevant to teams that want security checks embedded in existing developer workflows rather than a separate analyst-only console.²³

It is a weaker fit for buyers that require a long independent enterprise track record, Windows endpoint coverage today, or independently benchmarked detection quality before adoption. The paid tiers also become significant for teams with heavy private-repository or red-team usage.