An operator,
not an assistant.
Iris.Md started as a frustration. Every "AI assistant" we tried forgot what we taught it the moment the session ended, lived inside someone else's UI, and quietly billed us for the privilege of being held hostage.
So we built the opposite: a small, fast, open-source agent that lives on a machine you control, remembers what it learns, and grows more capable the longer it runs.
It speaks Telegram, Discord, Slack, the terminal — wherever you already are. It schedules its own work, spawns isolated subagents, and writes new skills for itself when it discovers a gap.
It's not an assistant. It's an operator. The difference is who is waiting for whom.
What we won't compromise on.
The agent belongs to you
Self-host first. Bring your own keys, your own models, your own infrastructure. We refuse to be a vendor you can't leave.
Memory is the product
Most agents forget their last sentence. Iris.Md remembers projects, decisions, and the shape of how you work — and gets sharper the longer it runs.
Operator, not assistant
It doesn't wait to be asked. It schedules, it watches, it acts. Assistants reply. Operators ship.
Boring on purpose
Rust runtime. Python skills. NATS. Postgres. Docker. Nothing exotic. The fewer surprises in the stack, the more attention left for the work.
Open in the dull, slow way
MIT license, public roadmap, signed-skill registry. No closed-source plugin economy. No telemetry by default.