Revery

The AI-wiki idea works. Running one by hand doesn't.

Teams that give their AI agents a maintained knowledge base get noticeably better results; that's the whole "company brain" argument. But the do-it-yourself version means managing markdown files, Git, custom prompts, OAuth plumbing, and brittle automation, and it still can't answer the questions a team actually has: who approved this fact, is it stale, does it conflict with what we decided last month, and who's allowed to see it.

Revery is that layer, run properly. Sources go in (documents, notes, interview transcripts). AI proposes what's worth remembering. A human reviews before anything becomes trusted memory. Every fact keeps its source, its reviewer, and its sensitivity level.

Then the whole team, and each person's AI agents, work from it. Agents connect over MCP with scoped, time-limited access: they can read, search, and propose, and they can't approve, edit, or delete. Agents propose; humans confirm.

Part of a pipeline

Revery is the destination of the capture-manage-productize method. Tacit captures expert knowledge through guided interviews. Revery turns it into shared, governed memory your team and their agents run on. For teams whose data can't live in someone else's cloud, Recall is the self-hosted edition of the same engine.

If you want the outcome before the product ships, that's what my Knowledge Capture engagements deliver today.