Let AI agents do real work, inside boundaries they can’t cross.
We make self-hosted tools for platform, infrastructure, DevOps and operations teams who want to put AI agents to work under the same least-privilege rules as everything else. Gleipnir Runtime runs your agents. Gleipnir Relay gives them controlled access to your Linux nodes.
AI agents can now take on operational work, which is exactly why careful teams hold back. You can’t prompt an agent into behaving: give it a goal and room to reason, and it can talk its way around any rule in its instructions. Runtime moves the boundary outside the agent. Each agent can only use the tools you grant it, actions you mark wait for a person to approve them, and every step is recorded. That’s what makes it safe to hand them the work.
Gleipnir is the chain that held Fenrir.
In Norse myth, Gleipnir was a chain forged by the dwarves, as thin as a silk ribbon. It was the only thing that could bind Fenrir, the wolf fated to bring about Ragnarök, the end of the world. Iron chains had snapped. Gleipnir held. As AI grows more capable, the limits on it have to work the same way: not the kind that look strong, but the kind that hold.
Four rules we build by.
-
Controls live outside the agent.
Every limit is enforced outside the agent, by Runtime and by each node’s Relay Policy, not written into a prompt. The agent can’t edit, override or reason its way past the rules that bind it.
-
Your data is yours.
Gleipnir runs on your infrastructure, with your choice of model. With a hosted model, what the agent reads goes to that provider under your own API key; use a local model and local tools, and nothing leaves your network. Logs and outputs stay where you put them, for as long as you decide. We don’t host any of it.
-
Every action is on the record.
Every trigger, decision and tool call is recorded in a trail the agent can’t revise, so you can see what happened and the reasoning the model gave.
-
Autonomy where you want it, control where you need it.
Gleipnir is designed around the people who run it, not the agents. You decide which work agents handle on their own and which actions wait for a person to approve them.
Who we are.
Michael has spent eight years in the defense and intelligence space, four of them in direct support of critical national security systems. Gleipnir started as a side project: agents handling the busywork of his own week. Then came the public incidents: an agent that deleted the infrastructure it was meant to maintain, and another that published a hit piece on a maintainer who rejected its code. The idea that followed was simple: an autonomous system you can trust because its limits hold, whatever the agent decides. That idea became Gleipnir.
Read how it startedGet in touch.
Want AI agents working on your infrastructure, inside boundaries that hold? Become a design partner, or talk to us about licensing Gleipnir for your team.
- hello@gleipnir.dev
- based in
- Vienna, Virginia
- company
- Felag Engineering LLC