A workflow control plane for AI agents.
Declared steps. Hard gates between them. Runs you can prove — who did what, what was checked, who signed off. The engine runs on the agent host you already use; no LLM API key, no cloud.
$ npm install -g riglane
A step cannot be stamped complete unless its gate actually passed — schema checks on every handoff, script verdicts, human sign-off. The rules live in the engine, not in the prompt.
Every step, tool call, gate verdict and retry lands in a trace you can open and audit. State is on disk: runs survive crashes, restarts and closed terminals.
A decision waits in the Inbox for hours or days without losing state — the agent proposes, the engine holds, the person disposes. All three leave evidence.
The catalog shares workflows as pointers with byte-locked capability inventories — you see every command a workflow can run before you install it.
The Workflow Studio, loaded with a real twelve-role research workflow — frozen mid-run, waiting on a human signature. Click through the graph, the live trace and the Inbox.