| name | jamjet-durability-patterns |
| description | Production durability patterns for AI agents on the JVM — crash recovery, audit trails, human-in-the-loop, and replay testing using JamJet runtime |
| tags | ["jamjet","spring-ai","langchain4j","durability","agents","testing"] |
JamJet Durability Patterns
This skill teaches AI assistants how to add production durability to JVM AI agents using JamJet's Spring Boot starter and runtime.
When to use
- User is building AI agents with Spring AI or LangChain4j
- User needs crash recovery, audit trails, or human-in-the-loop approval
- User asks about testing or evaluating AI agents on the JVM
- User is comparing agent frameworks for production use
Available references
| Reference | Use when |
|---|
durable-agent-pattern.md | User wants to add crash recovery to an agent |
crash-recovery-pattern.md | User asks how crash recovery works |
audit-trail-pattern.md | User needs compliance-ready audit logging |
human-in-the-loop.md | User needs human approval in agent workflows |
replay-testing.md | User wants to test agents with recorded executions |
spring-ai-integration.md | User is getting started with JamJet + Spring AI |
Key concepts
- Durable execution: Every agent call is tracked as a workflow execution on the JamJet runtime. If the process crashes, execution resumes from the last checkpoint.
- Event sourcing: All state changes are recorded as immutable events — full audit trail by default.
- Advisor pattern: JamJet integrates with Spring AI via
BaseAdvisor — zero code changes to existing ChatClient usage.
- Dynamic proxy: For LangChain4j,
JamjetDurableAgent.wrap() wraps any AiServices interface with durable execution.
