Problem
Our "agents" architecture is struggling from load on one server.
Mostly, we need to support Apps and Tools, outside the context of agents
Outcome
We will repurpose the agent server to become the Seed App Environment (SAE)
It will support:
Scheduling (Triggers)
Code Execution (Tools, Runs)
App Registration + Auth
We will move to 3 machines:
SAE dev+staging
SAE prod
Docker services (web search and markdown web reads)
2 new resource types:
App
Tool
An app is launched/running/mounted if its server claims to be running it. The server is defined by the siteURL of the space, or by the server URL as defined on the app mount.
A tool can be executed if it is the child of an app and if the app has been launched.
Open Questions
How to authenticate with SAE?
The app server should not allow anybody to run/execute apps. We need a permissions system, building upon the new Keyline architecture.
"Who can Run?"
1 Space/Owner Based
The owner can grant caps in a similar way to resource grants.
2 Space Writer Allowance - CURRENT BEST CHOICE
Everyone who can write to the site's space can run
3 Anyone can Run - AGENT SERVER DOES THIS TODAY
Allow anyone to run any app, even apps on different spaces
Migration/Adoption Plan
What is the order of operations to get to the new architecure
How will we support self-hosted architecture, as well as EE and the multi-tenant
Walking back (deleting) some features like space-level and app-level config of agent servers
Rabbit Holes
a
Scope
a
Do you like what you are reading? Subscribe to receive updates.
Unsubscribe anytime