Comparisons

Little Workflow is an AI workflow compiler plus a durable runtime plus a capability allowlist (a full capability firewall is planned after the alpha). That combination doesn't have a clean one-to-one analogue, so it's easier to define by contrast.

Versus Temporal

Temporal is a durable execution engine for human-authored workflows. Developers write activities and workflow code in their chosen language and Temporal handles retries, timers, and replay through history.

Little Workflow uses the same shape (durable state, replay from history) but the workflow graph is AI-generated from a declarative WorkflowDefinition, not hand-written. Temporal does not include a planner, a capability allowlist, or a model-call boundary — those are the parts Little Workflow exists for. Use Temporal when humans design the graph; use Little Workflow when the developer supplies a goal, capabilities, and policy, and wants the graph compiled and locked.

Versus LangGraph

LangGraph is an in-process graph framework for building agent loops in Python or TypeScript. Developers wire nodes, edges, and state explicitly.

Little Workflow does not try to be a node graph library you compose in code. The IR (LWIR) is a compiled artifact, not an authoring surface. LangGraph runs in your process; Little Workflow runs durably against a local, replayable event log with content-addressed WorkflowVersions. If the goal is "give me a thin agent loop inside one service," LangGraph is closer. If the goal is "the workflow must survive crashes, be replayable, and pass an audit," Little Workflow is the shape.

Versus n8n

n8n is a low-code automation builder with a visual node editor and a large connector catalog. It targets operators who wire integrations together by hand.

Little Workflow is not low-code first. The authoring surface is TypeScript and JSON. Visual editing is planned for v1.0.0, but the core thesis is that the workflow itself is compiled from intent rather than dragged together. n8n connectors and Little Workflow capabilities also differ in posture: n8n grants the workflow whatever the connector allows; Little Workflow lists the models, tools, skills, and code access a workflow may use, and the planner sees only the tools the definition makes visible to it. The full capability firewall is post-alpha.

When Little Workflow Is Not The Right Tool

• A chat UI with no durable workflow underneath. Use the AI SDK directly.
• An ad-hoc script that runs once. Use a script.
• A sub-second interactive loop. The durability tax is not worth it.
• A workflow whose graph never changes and was designed by a human. Use Temporal or write the graph by hand.
• A multi-agent research framework with no policy or budget enforcement. Use a graph framework.

Little Workflow earns its weight when the workflow is generated, the run is long-lived, the side effects matter, and someone will eventually ask "what exactly did the model decide, and why."