Building Custom Security Agents with Pydantic AI

About this talk

While popular coding agents excel at building security tools for SOC and threat hunting, they trap teams in single-vendor ecosystems where model refusals can disrupt defense workflows. Security engineers need the freedom to swap models—Anthropic, OpenAI, Gemini, or local open source—and correlate across multiple SIEM/EDR tools without vendor lock-in. This talk demonstrates how to build vendor-agnostic blue-team agents using frameworks like Pydantic AI, paired with OpenTelemetry for deep observability, to optimize development, reliability, and runtime performance.

The speaker

• MF

  Matthew Franz

Where it is