Evidence can disagree. It still stays in the case.

Write the hypothesis before the evidence arrives. Put each interview note, funnel number, and support ticket in the section that tests it.

A whole loop on one canvas: the hypothesis, the section of tasks it produced, and the evidence and insight linked underneath.

Bring your agents into the loop

Connect Claude Code, Codex, Cursor, OpenCode, or Claude on the web through Model Context Protocol. They work directly in a Meridian loop.

Agents stay inside your access

Prima works with your access and nothing more. She cannot open a private loop you are not in. Renaming or deleting a loop stays human-only.

Your workspace stays yours

Workspace boundaries are enforced in the database on every request. Work from another workspace is never returned.

Prima starts with the loop

Ask a question and Prima reads the loop before she answers. Her work goes back onto the same canvas under her own name, where the team can inspect it.

Prima's chat panel docked over the canvas, showing a tool step reading the loop's canvas context and a drafted status update citing the evidence on screen.

Every open question has a state

See every loop as cards or a list, with its status, due date, and how much evidence and insight it holds. Sort by due date or last update, and filter to your loops or the ones shared with you.

The loops table listing every loop with a status pill, owner and due date.

Prima tells you what needs you

Prima notices a section with no evidence after two weeks, or one with every task done and no verdict, and offers to act. Her nudges show on the canvas and in Needs you on your home page.

The signed-in home screen listing three loops with their status, evidence and insight counts.

Every Prima chat stays with its loop

Return to the thread that shaped the work without searching a general-purpose chat history. Each conversation names the loop it belongs to.

The chat index, where each conversation with Prima names the loop it belongs to.