A JSON-based workflow language where AI agents define, compile, and execute reliable and consistent workflows.
Purpose-built for agents to author, not just execute. An agent solves a task once, captures the logic as a validated workflow, and that workflow runs deterministically from that point forward — no re-prompting, no drift.
Tool calls, branching, and data flow execute with fixed logic — no prompt drift, no hallucinated detours. LLM intelligence is scoped to the steps that need it, so the rest of your workflow behaves the same way every time.
Validate the control-flow, data-flow, and types before the workflow is ever deployed. Through careful static analysis, RemoraFlow catches broken references, type mismatches, and unreachable steps at compile time, eliminating runtime surprises and wasted LLM calls.
The compiler determines which tool parameters are static vs. dynamic, enabling human supervisors to review and approve a limited set of behaviors ahead of time.
Block dangerous actions, require manager sign-off for sensitive ones, and let the safe ones through — with built-in approval routing, timeouts, and a full audit trail. Design flexible, cascading policies that every tool call passes through before it executes.
Plug in your favorite durable execution environment and workflows survive restarts, sleep across deployments, and resume exactly where they left off, all without wasting a penny. No framework yet? It runs just as well in a single process out of the box.