Blekline started from a simple observation: the teams we talked to weren’t worried about AI being too powerful. They were worried about losing control of it quietly, a customer detail pasted for context, credentials in a debugging session, destructive SQL after prompt injection.

Nobody making a mistake on purpose; just people moving fast in a world where the model is always listening.Audit logs catch it too late. Static IAM does not observe what agents actually do. Neither scales.So we built what was missing: open-core NHIM and runtime enforcement at the MCP, SDK, and ingress boundary. Mask, enforce, lineage, and metadata-only audit before execution. Not a dashboard you check on Fridays. Something that runs every time an agent acts.Open core: server, proxy, SDK, and cursor-hooks on GitHub and npm.
‍
Cloud: fleet policy, authoritative PII mask, Trust Vault, Lineage, SIEM, and enterprise residency options.

Today: Digital proxy and protocol layer. EEA-first by default. Enterprises should not have to choose between moving fast with AI and keeping non-human identities under control.