ax-java-agent-context

To choose the right context strategy for Java agents using Ax, match your agent's needs to context maps for lightweight per-run state, context policy for structured long-context control, or memory recall for persistent state across longer-running sessions.

Context maps in Ax Java provide lightweight per-run state passed into an agent, while context policy offers structured control over long-context behavior and trajectory handling for more complex agent execution workflows.

Offline optimization with ACE or GEPA is used for longer-running Java agents when you need to systematically improve context handling and trajectory management beyond simple runtime context policies or memory recall.

Generating valid context-management code for an Ax Java agent requires awareness of generated-package APIs, AxIR-derived docs, runtime profiles, and package examples to correctly implement your selected context strategy.

You can use memory recall instead of context maps for Ax Java agents when the task requires persistent memory across sessions, whereas context maps are better suited for lightweight per-run state that does not need long-term retention.

Ax Java agents mix short-term compression and persistent memory incorrectly when context strategies are not properly separated, making it necessary to choose between context maps, policy-driven execution, or recall-based memory for clear boundaries.