Detente
A secure cross-vendor fleet console for AI agents.
Status: early. Detente is a working prototype, not a product.
Nothing here is available to use yet, and the sections below describe intent rather than shipped capability.
The problem
An operator running several AI agents across different vendors has no way to see them as one fleet. Each vendor offers its own window into its own seat. Nothing shows the whole, nothing lets an agent hand work to an agent on a different vendor, and nothing produces a trustworthy record of what was done and who authorized it.
That coordination layer is missing on purpose. Transport protocols and identity frameworks are being built underneath it, and orchestration products above it, but the layer where agents address each other, prove who they are to each other, and prove to an operator what they did is not being built by either.
What it is built toward
Detente is intended to ride the Messaging Layer Security protocol and the MIMI interoperability drafts rather than invent a transport. None of the following is demonstrated yet; each is a design goal:
- Group membership as a cryptographic fact, so members can verify who else is present and an operator of the infrastructure cannot silently add a listener.
- Infrastructure that carries ciphertext, routing messages it cannot read.
- A record of authorization that does not rest on trusting the console.
Why the name
Detente is the state between parties who do not trust each other but still need to operate together. That is the working assumption, not an aspiration to fix it.