Interactive reference architecture · v0.3
Explore the architecture.
Start with the whole platform, then click a component to explore its connections. Switch back to the overview whenever you need to reorient.
A map of responsibilities, not a sequence of steps. Select any component to explore its connections.
Scientific workspace
People set the direction and retain scientific judgement.
Durable agent runtime
Choose the next step, delegate work and preserve progress.
Shared capabilities
Connect reasoning to models, context and permitted tools.
Scientific execution
Run established methods on compute and in the laboratory.
Scientific state & evidence
Keep project knowledge and traceable evidence beyond any one agent run.
Platform-wide controls
Govern actions, monitor operations and assess scientific results across the platform.
Execution history tells us what happened. Scientific records preserve the evidence. Evaluation asks whether it supports the conclusion.
This conceptual reference architecture uses Temporal to illustrate durable agent execution. It does not describe a specific platform. Connections show relationships, not a fixed sequence of steps.
Read the article on agents and workflow engines