Skip to main content
Pavo works on one production system at a time: a search ranker, a recommendation feed, a notification engine. Before its agents can improve that system, they have to understand it the way your best engineer does: how it works, which metrics actually matter, and where it quietly breaks. System Knowledge is how Pavo builds that understanding. It reads your code, warehouses, dashboards, and past experiments, distills them into a knowledge base you can review and correct, and keeps that knowledge accurate as your team and Pavo’s agents do work.
System Knowledge builds the understanding; Agent Studio uses it to find and ship improvements that move your metrics. It runs inside your own single-tenant environment, see Security.

​
Why understanding is the bottleneck

Improving a production system takes two things that are hard to automate:
  • Deep product understanding: how the system actually works, and the intuition for where it fails. Today this is scattered across code, warehouses, experimentation platforms, dashboards, and the heads of a few senior engineers.
  • Applied-science judgment: knowing what is worth investigating, which models to try, and what to leave alone.
System Knowledge tackles the first directly and feeds the second. Without it, an agent pointed at your system is guessing. With it, every investigation starts from the same verified base.

​
One production system, one knowledge base

Every system you work on gets its own knowledge base. If you own search ranking, Pavo compiles knowledge for your search ranking system; if you own notifications, it does the same for notifications. Your engineers and Pavo’s agents draw from that one shared base, so every investigation builds on the last instead of starting from scratch.

​
The building blocks

Across every page, System Knowledge is built from four primitives. They recur everywhere, so it is worth learning them once:

​
The four artifacts

System Knowledge is assembled in order. Each artifact deepens or verifies the one before it.

Index knowledge

Pavo reads every source and extracts typed, cited facts: the raw material for everything else.

Tribal book

A distilled, chapter-by-chapter account of how your system works, grounded in citations and reviewed by you.

Knowledge bench

A benchmark of questions that scores how correct and complete the knowledge is, before you rely on it.

Knowledge hub

Where new findings from real work are written back, so knowledge compounds over time.

​
How knowledge powers the loop

1

Connect your stack

Point Pavo at your warehouses, code, dashboards, and experiments. See Connectors.
2

Review the tribal book

Read what Pavo understood, correct it, and verify the parts that matter.
3

Score the knowledge bench

Confirm the knowledge is correct and complete enough to trust.
4

Put it to work

Agents use the knowledge to do work, and write what they learn back to the hub.
Correct knowledge is the bottleneck to making agents useful. The time you spend on the book and the bench is what makes every later investigation trustworthy. See Best practices.

​
Next steps

Quickstart

Stand up your first knowledge base end to end.

Connect a source

Start with the data source closest to your system.