Durable Execution
Persisting agent state at every step so runs survive crashes, pause for approval and resume later — a core pillar of Akashi.
Durable execution means a long-running workflow checkpoints its state so it can be suspended, survive process restarts and resume exactly where it left off. For agents it is what makes multi-turn memory, long tasks and Human-in-the-Loop approvals reliable.
How I use it
- Akashi — durability is one of the framework's pillars: versioned message serialization, a SQLite checkpoint store, and runs that suspend on an approval and resume in another process.
- langgraph-langserve-starter — SQLite checkpoints let clients resume a research thread over HTTP by
thread_id. - local-rag-chatbot — SQLite-backed chat history through LangGraph checkpoints.
- Conversational AI Agents — LangGraph.js with PostgreSQL checkpointing in my Independent AI & Full-Stack Engineer (Freelance) work.
Why it matters
Without durability, an agent that waits hours for an approval or crashes mid-task loses everything; with it, long tasks become restartable and auditable.