Updates · August 29, 2026
Why we started
The moment a bot keeps a job instead of just answering a question, you are not using an app any more, you are running a small workplace. And a workplace raises questions a chat window never did: who owns the computer, who is allowed to read the records, and what the bill is doing while everyone is asleep.
Grok Bot made a standing roster ordinary. You give a bot a job and it keeps the job. That's a different product from a chat session you drive yourself, even one that can loop and schedule if you already know the tricks. The default picture is an ongoing team: roles, responsibilities, work that continues while you sleep.
If it's a team, other humans have to be able to walk into the same room. The bots, the files, and the work belong to an organization. Chatticus is that workplace. v1 is one household. Every record already carries a tenant_id, so more people can arrive later without a rewrite.
A serious team also needs control.
The model is a choice. v1 talks to OpenAI, and Amazon Bedrock is next. We want stronger models, cheaper models, and local models — the workplace is the product, and the model is a part we expect to swap. We won't add a second client until the first path is honest.
The meter is a choice. A research job that kicks off at 4am doesn't need the expensive lane. If nobody is waiting, pay for patience: Flex, Batch, start earlier. Same prompt, same tools, same model; only the clock and the tier change. That argument lives in Cost versus SLA so we can't quietly default every turn to Fast.
The computer is a choice. Chatticus runs a real Linux workplace — the same Ubuntu image on Fargate, EC2, and local Docker. A bot should read files without waiting for a desktop, and a turn should pick a screen, a container, or a background job without changing the files. Snapshots are how hosts share a disk today. A single network filesystem is the direction when a VM, a container, and a job have to share one live tree. See Computer manifold.
The account is a choice. The workplace runs in AWS the customer controls. Privacy, logs, and spend are visible. If the work is serious, the processor is a contract you can actually read.
The work has to be visible. Channels are the conversation: every bot on a channel reads it; only the addressed bot acts. Tasks are the board across the whole roster. People and bots should share both.
That's the first-day bet. You can check what we wrote down: the product model, the source, and the specifications that spell out each control. The motivation is on this page; the rules are in the repository.