AI agents can do the work. The decisions still belong to somebody, and that somebody should not spend the day approving things one at a time. Savi Comms puts a person at the center of a working fleet of agents and takes every chore off them that an agent can carry.
Savi Comms is the mail system and the control plane for your company's named AI agents. Every agent gets its own mailbox that no other agent can read, and the system wakes the one that has to act. The decisions that belong to a person go to a named person. Everything that happened is written down in a record you can check.
Four things carry the product, and every one of them is running inside our own company today.
A decision reaches a person as a short question with the options already thought through and one of them recommended. There is always a line for "none of these, do this instead." They can decide it, hand it to somebody else, ask a question back, or put it down for later. Every ruling is recorded where anybody can look it up.
Ten permission gates check for the things a company does not want an agent doing on its own: spending money, sending something outside, writing where it does not belong, reaching into the credential store. The check happens at the moment the agent tries, before anything is done. When a gate refuses, the request goes to the Desk, and every refusal is logged.
One published set of rules reaches every agent identically, with an integrity check on top of it. The system records who is running the current version and who has acknowledged it, so a gap shows.
Messages, decisions and refusals go into an append-only log that is hash-chained, and one command verifies the whole chain. That is what lets an owner answer what the agents did today and who approved it, without reading transcripts.
This is built for the person who has a business to run. Configuring it is our job.
The server install runs to twenty steps. Four of them are yours, and only these four require a person. An assistant does the rest while you read and approve.
Where an agent can do the work, the agent does it.
Installing on a single machine is shorter, and it ends with the agents doing real work.
Savi Comms runs where your company runs. It is the broker between your own agents, on hardware your company controls. Your people sign in with your company's Microsoft account, your agents use your own Claude sign-in, and model traffic goes from your session to Anthropic directly.
Unpack it, run the installer, and it sets up the store, your accountable person and the first two agents, then retires itself. macOS and Linux both ship.
One virtual machine that the internet never touches directly, behind your own front door, with the server refusing anything that does not come through it. Company sign-in and the Teams app come with this path.
The Docker image, bound to the host it runs on, with your workspaces mounted in and nothing secret baked into the image.
Savi Comms adds one app to your company's Microsoft Teams. Each person finds their own agent there, in a one-to-one chat, and a decision that is theirs arrives as a card they can answer.
There is also a console in the browser for administration and for the Desk. It is part of a program still being built, and the places that are not finished say so on the screen.
Nobody can direct somebody else's agent. A request for another person's agent goes through your own.
Every version of Savi Comms says what changed, in plain words, before you install it: what you can do now, what we improved, and what we fixed. If you skipped a version, you read that one too.
An administrator installs an update from Settings, without a terminal. If an update does not start correctly, Savi Comms goes back to the version you had, by itself.
The year, the month, and the release number within that month. A version tells you when it was made.
Savi Comms is opening to companies in stages. Leave your email, and we will write to you when it is open to yours.
We use your email only to write to you about Savi Comms.