The agent stack, floor by floor.
Every AI agent runs on the same layers: model, provider, transport, runtime, platform, and surface. Walk them as one building, drawn live from the Atlas graph.
Every room is a live record in the a5c Atlas graph, not a screenshot.
Interactive: click a floor to step in, a room to enter, a desk to focus one record; scroll to zoom, Esc steps back. Records link into the live graph.
01The layers
One stack, three floors.
Every agent, ours or anyone’s, is assembled from the same layers. The tower is a projection of that stack from Atlas, our public knowledge graph of the AI-agent ecosystem. It is the map our harnesses are calibrated against.
- Base layers
- Model, provider, transport, and agent core: the artifact, the endpoint that serves it, how calls move, and the loop that turns a prompt into steps.
- Runtime
- Runtime, platform, workspace, and sandbox: where the loop runs, what orchestrates it, the tree it works in, and the invariants that bound it.
- Surfaces
- Products, UI, capabilities, and plugins: the user-facing agents built on the stack, and the extension points that add skills to a harness.
We calibrate a harness to your domain across this whole stack, then operate it with your team.
The tower shows the terrain; the solution is a harness tuned to your corner of it and run until it holds.