Our go-to-market runs out of KnowDrive

Every company that sells agent infrastructure says they use their own product. Most mean they tried it once. We mean the thing that plans our week cannot function without it, and when the store is down, the work stops.

The loop

The runner is unremarkable, which is the point. It reads its state out of a KnowDrive store, does one unit of work, and writes the result back as a new document in the same store. The next session starts by reading what the last one wrote. There is no separate database, no state file in a repo, and no prompt that gets longer every week.

• Session handoffs — what was decided, what is open, what comes next — live as documents, not as chat history.
• Memory sync mirrors durable facts into the same corpus, so a fact learned in one session is retrievable in every later one.
• Because it is a corpus and not a context window, the tenth session costs the same to start as the first.

The interesting property is not that the agent remembers. It is that the memory outlives the model we happened to be using that month.

What this actually buys

The obvious benefit is that no one re-pastes context. The real benefit is narrower and more useful: when we change models, nothing migrates. The corpus was cut once, sized for the smallest reader we might ever point at it, so switching the reader is a configuration change rather than a re-embedding project.

That is the whole VORAS argument, and running the company this way is the only reason we believe it rather than merely assert it.

Where it is still rough

Retrieval quality is only as good as what you wrote down, and a bad handoff poisons the next three sessions. We have had to get disciplined about writing the handoff as if a stranger will read it, because in effect one does. That discipline is a cost, and it is worth naming rather than hiding.