The problem we kept hitting is that there's no "multiplayer mode" for assistants, that share project knowledge.
Today we're rolling out Estratos at https://estratos.ai
Estratos gives teams shared memory layers for their AI assistants.
* Stacked memories *
The layers part may be a bit specific to us, so I'm very keen to get feedback.
The way we work, we set up the "house rules" for all Cowork projects at the memory root level. Then each project under the root has its own memories, and each subproject (eg. gtm) its own memories.
In the Cowork project instructions, we tell the agent to read memory from eg. root.project-acme.go-to-market and it'll bundle everything up the stack as the initial context for that Cowork session.
* Revision control *
The product has a distinction between contributing users and admins. Users can propose memories and updates (and use whatever they propose), but these changes aren't visible to the rest of the team until admins approve them.
* First-use user experience *
We're aiming this for non-technical users and tried to keep the installation method as simple as we could think of: a line to be configured in eg Claude Desktop UI settings, and the normal OAuth authorisation process as it configures the MCP server for Estratos. We're already thinking about extending this delivery method to allow shared skills.
* Pricing *
For now we're keeping it free. We're leaning heavily towards charging $2/mo per seat, but nothing settled yet. It'd be great to get some feedback on this too.
Thanks in advance.