membrane labs · private beta

agents act fast. brane decides first.

Membrane Labs at membranelabs.org builds the control layer for AI agents. Brane — policy-as-code that runs before every model call, tool call, and output.

membrane labs · membranelabs.org

Membrane Labs is an AI agent infrastructure lab building Brane for runtime policy control, AI agent guardrails, tool call governance, and MCP tool governance.

the control plane missing from every ai stack.

What we believe

Agents are moving from demos into production systems. The piece they're missing is an enforcement layer — the layer that decides what an agent is allowed to do before it does it.

Read the Agent Mesh thesis

What we write

Engineering notes on agent policy, retrieval, memory, routing, and the runtime boundaries where agents become real software.

Browse Lab Notes