Memorylayer turns Engram into something agents can actually connect to: workspaces, API keys, a hosted bridge, bootstrap payloads, and a dashboard that keeps the runtime usable.
Long-lived Python service and Postgres, instead of forcing memory through a serverless shape.
Every workspace exposes API URLs, tool discovery, and starter skills for plug-and-play setup.
The goal is making hosted memory usable, not gatekeeping it behind fake plans.
The core package stays focused on memory behavior. Memorylayer handles the operational surface around it.
Create workspaces, inspect memory health, issue keys, invite collaborators, and keep audit history in one place.
Use hosted endpoints for search, remember, recent memories, audit history, bootstrap, and a compact HTTP bridge for selected Engram tools.
This is still the real Engram engine underneath, so retrieval, graph, and memory improvements pass straight through to the hosted layer.
The service is shaped for real agent integration, not just for browsing a dashboard.
Each workspace exposes a bootstrap payload with API URLs, bridge URLs, and starter skills for Codex-style setups.
Agents can work through the service instead of assuming they have local access to the underlying Engram runtime.
It already has the right shape for dedicated deployments, usage controls, and future agent session capture.