The map of the kits — which sits on which, and what each one is for

You are here: Building with agents › The kits

Summary#

You never add 29 kits. You add the two or three your shape takes, and the bridges between them arrive by themselves.

Solid boxes are kits you add with one line. Dashed pills are bridges: nobody writes them in an app.

Description#

How do I read the map?#

• Left, bottom to top. Llm is every call to a model. Agents sits on it. Three kits draw screens on top of Agents. Under Llm are the providers and a router: you pick the ones you use.
• Right. Five kits that have nothing to do with AI. Each is useful in an app with no agent in it: a list of process runs, an approval in an inbox, a piece of work owned until it is done.
• Between. When an app uses a kit from the left and a kit from the right, a bridge connects them. A bridge is a small kit of its own. It is not written in the app: osy lock adds it and says that it did. Neither of the two kits it joins knows about the other.

That last point is why the family comes apart cleanly. An app with agents and no business processes carries no process code. An app that adds Processes later gets the agents-and-processes bridge the same day, with nothing to wire.

Which kits does my shape take?#

What is each kit for?#

Models

Agents

People

Processes

Work that lasts

Privacy

Released with the family

What does a kit's page tell me?#

Every kit page has the same five parts:

1. What it is for, in a sentence, and how you add it.
2. What you get without writing anything.
3. When you need it, by shape.
4. What it exposes: its settings, the types you derive from, the policies you answer and what each admits until you do, the functions you call, the events you can hear, the slots you can fill, the screens you mount.
5. Go deeper: the reference pages behind it.

Part 4 is not written by hand. It is generated from what the tools say about the kit, so it cannot list a function that is not there. You can ask the same question yourself:

What to read next#

• Osysharp.Agents — an agent given a real job inside an app, and trusted with it — the core kit's page: start here to see what a kit page looks like
• One piece of work, followed end to end — one piece of work passing through six of these kits
• Deciding what to build — four shapes of app with agents in it — which kits your shape takes

Go deeper#

• Agents (a class with a job, its own account, a budget and a record) — the reference pages about agents, by subject
• The Llm kit — call a model, and know what each call cost — the Llm kit in full
• Processes — every run of a business process, listed and read from start to now (Processes kit) — the Processes kit in full
• The WorkQueues kit — queues, who holds what, and the service levels promised on any item — the WorkQueues kit in full
• The Privacy kit — erasure that keeps only what the company must or may — the Privacy kit in full