Companions
An AI teammate with a job, not a blank chat
A companion has its own instructions, its own memory, its own files, its own routines and its own model settings. It does not start from nothing every time you talk to it.
The problem
A fresh chat forgets that you exist
Every conversation with a general assistant starts by re-establishing who you are, what the company sells, which customers matter and how you like things written. That re-explaining is most of the cost of using AI at work, and it is paid again on every single thread.
- InstructionsHow it works, in its own words
- MemoryWhat it has learned, scoped to it
- FilesDocuments it produced and reuses
- ToolsExactly what it is allowed to call
What it does
Its own instructions
Written once, versioned, and published. Editing always creates the next draft rather than silently changing what is live.
Its own memory and files
Durable facts scoped to this companion alone, plus a file space it can write into mid-conversation and you can read afterwards.
Its own routines
Work it does on a schedule or on an event, without anyone opening a dashboard to start it.
How a companion behaves
Publishing is what makes a companion usable: only published ones appear in chat or can be given a routine. Memory is proposed rather than assumed.
- Saving something to memory shows you the diff first and never writes on its own
- A companion can be created from the Settings form or described in plain language in chat
- Scope can be personal, a department, or the whole organisation
The rest of it
Give your AI a company to work for
Connect your tools, and every agent you already use starts the day knowing how your company actually works.