Aeris is an open agent built by Aeris Research. Talk to it in the portal, or pull the runtime onto your own machine and let it operate inside your loop.
Beta is live · testers already in. One agent, everywhere you already are.
Hops into voice, listens, talks back, and coordinates the room in real time.
Live co-op partner — calls plays, watches your screen, learns your style mid-match.
Sits in the browser and actually does the task — not another tab that explains how.
Spin up agents with their own voice, vibe, and rules. Mee6, but it actually does stuff.
Aeris Research releases model weights, agent traces, and runtime code. We publish what we learn — including what doesn't work.
The AI industry is starving for high-fidelity human behaviour. Aeris turns gameplay into the dataset — every match, every call, every reaction structured into the pipeline.
Live signal
Voice, input, decisions, micro-reactions captured frame by frame.
Scale Opal pipeline
Sessions stream into structured behavioural datasets, ready for training.
Paid to play
Contributors earn from every qualifying session. Your time, your data, your cut.
Drop the CLI on any Unix box, Mac, or Windows machine. Same protocol as the hosted portal — your conversations, your tools, your loop.
$ pip install aeris-agent $ aeris login → connected to aeris-research · session active $ aeris chat > design an outreach plan for this week aeris: tracing context · synthesizing…
▍
The moment you let an agent act, you have no idea if it's actually working for you. Aeris is the missing ledger and spend control — built into the runtime, not bolted on.
$ aeris loop --explain
> We built Aeris for agents that actually do paid work.
It gives an Aeris agent a P&L. The agent earns, spends,
and keeps its own books — so it runs a business instead
of just calling tools.
> the loop:
[01] earn
It prices a job and sends a Stripe payment link.
Earning is fully autonomous.
[02] spend (gated)
To fulfill, it buys the APIs and data it needs
through Stripe. Every spend is gated by one human tap.
The agent cannot approve its own money.
That gate is the safety, not a limitation.
[03] fulfill (local)
The work runs on a local model. Anything touching
cards, customer data, or the ledger routes to the
on-box runtime and never leaves the machine.
[04] book + reprice
Every dollar in and out is booked double-entry.
The agent reads the book and reprices itself.
> who it's for:
Anyone running an agent that spends to do paid work.
The moment you let an agent spend, you have no idea
if it's profitable. Aeris is the missing ledger and
spend control.
> outcome:
An agent that earns more than it spends,
with every spend approved and booked.
▍ try it below
no spends queued. agent cannot move money without you.