Dynamic Graph

Inside a Graph, if an Agent is given the right tool permissions, it can create, clone, activate, or suspend other Agents. This allows the Graph to expand dynamically during execution.

Concretely, an Agent can instantiate a peer Agent from an existing preset, clone a running Agent — its config, context, and tools — into a fresh identity, or run an entire blueprint as a sub-Graph and act as its own human, answering the sub-Graph's questions itself. The built-in chat Assistant uses this to act as a router that stands up and drives specialist sub-Graph teams rather than doing everything alone.

This capability set — an Agent growing the Graph at runtime — is what we call Prolifera Engineering. A few invariants are worth stating plainly. A cloned or created Agent is born disconnected and inactive: it must be wired in and activated before it does anything. An Agent creates or clones only when it chooses to call the tool — a long context does not trigger it; that is handled separately by Meditation. And an Agent can do this at all only if its preset grants the founding tools (create_subgraph, clone_agent_node, create_agent_node, create_task_board). Because that decision is the Agent's own and is not globally capped, a granted Agent can fan out without limit; the authority is concentrated and observable by design, not prevented.