README
Agent Runtimes / examples/AgentLoopExample
examples/AgentLoopExample
AgentLoopExample - Define and launch an agent execution loop.
This example demonstrates the "loop" paradigm (observe → think → act →
evaluate) instead of one-shot prompting. It is fully driven by the generic
loop specifications defined in agentspecs (src/specs/loops.ts):
- Pick any loop spec (Data Analysis, Plan/Execute/Critic, OODA, Human-in-the-Loop). The list is read from the loop catalogue, so adding a new YAML loop spec automatically makes it available here.
- The agent's system prompt is composed generically from the selected loop's objective, phases, constraints, termination policy and human settings — no loop-specific code.
- The agent then runs the loop against a live Jupyter notebook, where the loop state (dataframes, charts, intermediate results) lives outside the model so each iteration stays small and focused.
To run this example:
- Start the agent-runtimes server.
- Select this example from the header dropdown (Agent group).
Functions
References
default
Renames and re-exports AgentLoopExample