Keryn and Semgrep
Semgrep is a static analysis platform with an open rule format that Keryn also reads.
Every row links the source it comes from (how these pages are written). Vendor statements quote the vendor’s own pages as read on 18 November 2026; products change, so follow the link before relying on a row. A missing row is not a claim that the vendor lacks something.
What they do
Section titled “What they do”| Claim | Source |
|---|---|
| Static analysis in “35+ languages”, with cross-file analysis. | Semgrep: supported languages |
| AI rules: “27 AI Security rules covering prompt injection, unrestricted tool use, and data exfiltration”, run inside agentic coding tools. | Semgrep: risks in AI-generated code |
| “Code is not uploaded. Only findings are sent to Semgrep AppSec Platform.” | Semgrep quickstart |
Where they are stronger
Section titled “Where they are stronger”| Claim | Source |
|---|---|
| More languages, more rules and a commercial cross-file engine. | Semgrep: supported languages |
| Assistant proposes fixes and filters likely false positives in pull requests. | Semgrep Assistant |
Where Keryn differs
Section titled “Where Keryn differs”| Claim | Source |
|---|---|
| Keryn reads Semgrep-format rules and returns the same 13,475 result rows as Semgrep CE over 164 cases, with 0 differences. | Semgrep CE parity receipt |
| Keryn adds edges where authority changes (model output into a shell, an agent into a tool) with a proof level on each finding. | Why the edges |
| Keryn keeps an AI-BOM of models, SDKs, agents, tools and MCP integrations, exported as CycloneDX. | Capabilities |
| Keryn’s general static analysis recall is low; its published measurement found 15 of 83 held-out fixes. | Limits |
Measured results for Keryn are on the evaluations page, with what each one does not show.