Manwe: The ontology layer for the AI you use

Manwe gives the AI you use a lasting understanding of your work. Keep people, projects, decisions and constraints connected in an ontology you can inspect and edit. Separate personal, company and client context into Worlds. Bring local models, Claude Code or Codex sign-ins, or your own API keys to the same workspace. Use advisors, skills and workflows with explicit access controls. Currently free for Apple silicon Macs. Persistent agents and learning are experimental.

Hey Product Hunt, I am Oncel, maker of Manwe.

Manwe is built on a simple conviction: the most expensive failures rarely come from a lack of answers. They come from invisible assumptions, collapsed context, false consensus, and decisions that were never forced to withstand pressure before the organization acted.

Most AI products optimize for response. Manwe optimizes for decision quality.

Every serious Manwe run begins by imposing structure on the problem. It builds an ontology of the decision: actors, constraints, incentives, claims, evidence, assumptions, scenarios, risks, watchpoints, and what would change the verdict. Then it pressure-tests that map through advisors, trials, dissent, and model teams.

The output is not just another answer. It is an accountable decision record: what was believed, what was contested, where the evidence was strong or weak, why the recommendation moved, what futures should be watched, and what action should happen next.

The web app is for fast decision records. The Mac beta goes deeper with Worlds, governed memory, What Manwe sees, scenario testing, local/cloud/custom models, and Diagnostic Trials, where a recommendation is put on trial instead of being allowed to sound confident in private.

The feedback I care about is direct:

Where does Manwe expose risk that ordinary AI chat hides?
Where does the decision record need to become sharper before operators would trust it for consequential work?

i read this twice, which i do not usually do on a launch page. the idea that a recommendation should be put on trial instead of just sounding confident is very close to a problem i deal with a lot. small question before i try the beta: at the start, how much of the structure am i expected to write, and how much does manwe build by itself? asking because that is usually where i give up on tools like this.

Hey Product Hunt, I am Oncel, maker of Manwe.

Most AI products optimize for response. Manwe optimizes for decision quality.

The feedback I care about is direct:

Where does Manwe expose risk that ordinary AI chat hides?
Where does the decision record need to become sharper before operators would trust it for consequential work?