Knowledge Supply Chain
Right now, somewhere in this loop, a fact is changing. Watch one.
- Somewhere in this loop, a fact is changing right now.
- A new fragment arrives: a changelog note reading, PostgreSQL 14 is no longer supported.
- The system reads it and recognises two things inside the sentence: PostgreSQL, and the version, 14.
- A human reviews it and marks it trustworthy. Nearby, a fact from a meeting — a Knowledge Session — gets the same review.
- The validated fact joins a corpus: a small, connected cluster of related facts about the same system.
- That corpus is one part of a much larger, connected whole: Graph Memory. An older fact, elsewhere in the graph, is now in question because of this new one.
- Activity converges on the Knowledge Compiler, which takes the connected facts and produces several different outputs at once.
- The outputs reach three different readers: a human gets documentation and support material, a developer gets OpenAPI, and an AI agent gets MCP, prompt recipes and runbooks — together, the Agent Kit. All of it passes through the same API.
- An agent tries to install the system on a real server, using the runbook it was given. Something doesn’t match, and the install fails.
- The failure itself becomes a new fact, and it travels back into the system instead of disappearing.
- It rejoins the same corpus it came from, slightly improving it. This is the Discovery / Publishing loop: the system doesn’t just store knowledge, it keeps it correct. The loop continues.