Shannon

Shannon is an AI pentester from Keygraph that tests web applications and their APIs with access to the source code. It reads the code to find attack vectors, then uses browser automation and command-line tools to run real exploits against the running application. A finding reaches the report only when Shannon produced a working proof of concept, and hypotheses it cannot prove are discarded.

A scan runs in stages. Reconnaissance explores the running application and ties its behavior back to the source, with specialized agents for injection, XSS, SSRF, authentication, and authorization. In parallel, agentic code analysis maps the architecture, trust boundaries, exposed interfaces, data flows, and high-risk assets. The two streams of candidates are merged and deduplicated into an exploitation queue, exploitation agents attempt proof-of-concept attacks, and only the demonstrated ones are reported. Each scan runs in an ephemeral Docker container that mounts the repository read-only, and results stay in a local workspace.

Shannon is meant for staging, development, and local environments with disposable data, at the cadence teams ship, alongside human pentesters rather than in place of them. Its exploitation agents can create users, submit forms, and change application state, so it is not for production systems. A full run takes roughly 1 to 1.5 hours. Findings still need human review.

Model requests go directly to the provider or endpoint you configure, and Keygraph never proxies them. Anthropic, OpenAI, xAI, and AWS Bedrock are built in, and other providers in the Pi harness catalog, routers, LLM gateways, and local models are supported too. Keygraph's commercial Pro and Enterprise editions run the same agent and add black-box testing, dependency and secret checks, findings management, and fix pull requests.