Based on my research, here are the latest uses of SQLite for durable workflows as of 2025-2026: ## Key Developments and Use Cases ### 1. **Cloudflare Durable Objects with SQLite (2024-2025)** Cloudflare introduced zero-latency SQLite storage in every Durable Object, enabling durable workflows with: - **Platform-managed durability** with automatic backups - **Tight integration with Workers** for serverless execution - **Regional pinning** with strong transactional semantics - **Smart Placement** to minimize latency to the primary database ### 2. **Edge Database Revolution (2025)** SQLite has become the foundation for edge databases with global replication: - **Turso/libSQL**: Managed service with read replicas near users, supporting multi-tenant provisioning and SQL compatibility - **Cloudflare D1**: Managed SQLite for Workers with platform-level durability - **LiteFS (Fly.io)**: User-space filesystem replicating SQLite via WAL shipping - **Litestream**: Continuous backup and disaster recovery for SQLite ### 3. **Durable Execution Engines** **Persistasaurus** (Java prototype by Gunnar Morling): - Sub-1000-line durable execution engine using SQLite - Code-defined workflows with persistent steps - Checkpoint-based recovery for long-running processes - Particularly useful for AI/LLM workflows where repeating expensive calls is undesirable ### 4. **Workflow Orchestration Tools** - **Dagu**: Local-first workflow engine with SQLite backend, Web UI, and AI agent integration - **py-orchestrate**: Python workflow orchestrator with SQLite persistence - **Sayiir**: Embeddable durable workflow engine in Rust with Python/Node.js bindings ### 5. **Production Patterns for Durable Workflows** **Single-Writer, Many-Readers Model**: - Primary region handles all writes - WAL shipping to read replicas globally - Read-your-writes consistency controls - Session stickiness for consistency **Replication Patterns**: 1. **Leader-Follower with WAL Shipping** 2. **Partitioned Primaries** (per-tenant or per-key) 3. **Write-Queue at Edge** with primary commit via Durable Objects 4. **Offline-First** with local SQLite + CRDT/OT merge ### 6. **AI and Agentic Workflows** SQLite is increasingly used for: - **LLM workflow durability**: Preventing expensive API call repetition - **Human-in-the-loop workflows**: State persistence for approval processes - **Multi-step AI agent orchestration**: Checkpointing for fault tolerance ### 7. **Cost and Performance Benefits** - **Reduced latency**: Single-digit millisecond reads vs. 150ms cross-continent - **Lower costs**: Fewer egress penalties, simplified operations - **Simpler operations**: No network hops, predictable per-request latency ## Limitations and Considerations - **Single-writer constraint**: Multi-region multi-writer is not SQLite's comfort zone - **Consistency trade-offs**: Requires careful design for global read-after-write - **Platform limitations**: Some features (like global read replicas) are still evolving ## Emerging Trends - **SQLite "eating the cloud"**: Moving state closer to compute - **Hybrid approaches**: Combining SQLite at edge with centralized databases - **AI-native workflows**: SQLite as the persistence layer for agentic systems The landscape shows SQLite evolving from "just for mobile" to a production-grade solution for global, low-latency durable workflows, particularly when combined with modern replication layers and edge runtimes.