Yash Thakker builds Beam, a local-first security console for AI coding agents. The work started from a practical problem: teams were handing Claude Code, Codex, and Cursor real credentials and real repositories, and had no record of what those agents actually did afterwards.
Beam's answer is deliberately narrow — record agent activity on the machine it happens on, redact secrets before anything touches disk, and flag risky patterns for a human rather than pretending to block them. The detection core is open source under AGPL-3.0.
Focus
- Agent observability — what a coding agent did, with the evidence attached, not just what it was asked to do.
- Pre-run review — scanning skills and MCP server configs for hidden instructions before an agent trusts them.
- Accountability for autonomous systems — the audit and policy layer teams need once more than one person is running agents.
Yash also founds and runs explainx.ai, where he has taught over 300,000 learners and built AI platforms used globally, and bgblur.com.