Julep

Julep is a serverless platform for building durable, composable AI agents and workflows that crash-resume, retry safely, and restrict tool access, backed by Temporal and PostgreSQL.

Julep is an open-source, serverless platform for building production-ready AI agents and workflows. Instead of ad-hoc loops, it models agents as composable, durable dataflows: flows that can crash and resume, retry safely, and explain every step through a derived projection, with strict control over which tools the model may call. Under the hood, Julep combines a core service for agent definitions and task execution, persistent storage backed by PostgreSQL with vector capabilities, and a Temporal-based workflow engine for reliable execution. Connectors integrate external services and tools, while a declarative YAML-based configuration lets teams define multi-step pipelines that scale, retry, and recover automatically. Because execution, scaling, and retries are handled for you, data and ML teams can iterate on complex AI operations without provisioning servers. Julep 3 has shipped as a release candidate in 2026, signaling active development and a focus on durability and safety for agentic workloads.