v0.7.0 — Adaptive execution

Adaptive execution

v0.6.0 made execution governed. v0.7.0 makes it adaptive — the graph now reacts to why a call failed and how much budget is left, instead of retrying blindly or stopping hard:

• Reason-aware RetryPolicy — a pluggable FailureClassifier sorts each failure into TRANSIENT (retry, honouring a Retry-After hint), PERMANENT (stop immediately — no wasted attempts), or OVER_BUDGET (surface an InterruptRequest for human approval). The default classifier already understands JDK I/O timeouts and Spring AI / Spring Web 5xx + 429 vs. other 4xx — with zero compile-time Spring dependency. Add your own rules via orElse(...); the legacy retryOn predicate still works unchanged.
• Budget-aware RoutingStrategy — RoutingStrategy.budgetAware(budget, scope, threshold, premium, fallback) routes to a premium agent while budget remains, then degrades gracefully to a cheaper fallback once the remaining budget at a scope drops below a threshold. Deterministic and free: it reads the live BudgetPolicy counters, with no extra LLM call to classify complexity.

Both build on the existing BudgetPolicy, RetryPolicy and graph machinery from v0.6.0. Defaults are unchanged, so upgrading is drop-in.

Docs

• Resilience & error handling — reason-aware retries + cost-aware routing

Install (JitPack)

<dependency>
    <groupId>com.github.datallmhub.agentflow4j</groupId>
    <artifactId>agentflow4j-starter</artifactId>
    <version>v0.7.0</version>
</dependency>