Clarity Labs builds deterministic AI agents for regulated industries, starting with commercial insurance.
CL-Pipelines is a runtime-agnostic TypeScript library for long-running jobs and agent primitives. It gives you a typed phase model, explicit checkpoint protocol, and built-in retry modes — the infrastructure you need when regulated work must survive crashes, resume from exactly where it failed, and report progress to end users.
Who it's for
Teams building workflows that cannot afford to restart from scratch on failure. If you're running multi-step jobs — document processing pipelines, approval chains, deterministic agent loops, human review flows — and you need durability, resumability, and visible progress, cl-pipelines is the layer that holds it together.
How it differs from a job queue
Tools like BullMQ, Inngest, or cloud queues handle scheduling and delivery. CL-Pipelines is the execution model inside a job: a first-class typed phase model, explicit Checkpoint<TState> protocol, per-phase retry modes (resume vs. full retry), and structured error types. You bring your own scheduler; cl-pipelines tells you exactly what ran, what failed, and what state to resume from.
How it differs from an agent framework
LangGraph and CrewAI are opinionated about infrastructure and graph topology. CL-Pipelines is runtime-agnostic: it runs wherever JavaScript runs, uses the AI SDK's provider-agnostic LanguageModelV2 interface, and doesn't prescribe a graph structure. The agent primitive is a thin loop on top of the same phase runner — checkpointed per turn, not per workflow — so teams can wrap model behavior in durable, inspectable execution.