Concepts

Patterns that compose catalog tools into a working AI stack.

Sort: Recently updated

Agent Systems

How the pieces of an agent fit together and stay reliable

Sandboxing — running code in an isolated environment.

· Doc version 1.0

Sandboxing

Running the code an agent writes or pulls in inside an isolated environment — stopping destructive commands, secret leaks, and runaway resource use while still proving the code actually works

sandbox isolation security containers
Tools — the functions that connect a model to the outside world.

· Doc version 1.0

Tools

The functions that connect a model to the outside world — web search, scraping, browser control, and app integration that bridge to real data and actions beyond the training cutoff.

tools tool-use web browser
Harness engineering — the scaffolding built around an LLM.

· Doc version 1.0

Harness Engineering

Wrapping an LLM in scaffolding — a control loop, tools, a sandbox, guardrails, and evaluation — so it behaves reliably in production.

harness reliability architecture