Harkaq
A cage for your company's AI agents: Claude Code and Codex work freely inside, nothing leaves unchecked.
The founders
The company
Harkaq is a security layer that boxes AI coding agents like Claude Code and Codex into disposable, isolated workspaces with no keys and no network of their own. Every action routes through the Harkaq Gateway, which checks it against project policy, injects real credentials only on the way out, blocks secret leakage and logs everything for the CISO.
AI coding agents can read code, use credentials and touch the internet — and a single poisoned README or web page can hijack them, which is exactly why security teams are blocking them outright. Harkaq's bet is that enterprises want these agents switched on, not banned, so they need a cage rather than a ban.
The company was founded in 2026 by Gonzalo Márquez and is building out of Peru, framing itself around a problem security teams already know well: the choice between developer speed and keeping agents from leaking secrets or acting outside policy.
With a single named founder and limited public pedigree, the proof will be in the product — but the founding team has zeroed in on a concrete, well-scoped problem: credential handling, policy enforcement and audit-ready logging for agent sessions.
Who else is building in this space
Companies working on the same thing, at every stage.
Scale-up
Enterprise AI security and orchestration platform
Enterprise AI agent identity and access management.
Runtime authorization and control for AI agents.
Workforce AI security platform that discovers shadow AI, prevents data leakage, and enforces governance across enterprise tools.
Independent security testing for AI-built applications.
Secure AI gateway: one API for many LLMs with identity, policy and threat controls.
Thousands more founded every month.
Dealroom tracks new companies as they appear — founders, funding and signals, over the API or in the platform.