Flower Agent¶
Flower Agent lets you build and run agentic applications on Flower. An
AgentApp contains your agent logic, while Flower supplies model access,
connectors, run state, and a runtime that can execute the app locally or on
SuperGrid.
Choose how to start¶
Start with whichever interface you prefer:
Chat in your browser gets you to a first response without a local installation or model API key.
Chat in your terminal introduces the
flwr chatterminal interface.
When you’re ready to build:
Write your first AgentApp builds the smallest useful custom app.
Build a collaborative research agent adds conversation context, multiple connector calls, and a bounded tool loop.
Note
Flower Agent is experimental. Its interfaces and behavior may change between releases.
Understand the pieces¶
An AgentApp is packaged agent logic that Flower can execute.
An agent is an AgentApp made available for selection by an app spec such as
@flwrlabs/flwr-agentor by a specific Flower App Bundle (FAB).A run is one execution. Related runs can share a run series, which the chat interfaces present as a conversation.
A federation is the SuperGrid workspace in which runs execute and are visible.
A connector gives an AgentApp a runtime-provided tool. Some connectors are built in; account connectors use access granted by the signed-in user.
Read Use agents and federations to see how these pieces fit together and how to select agents and federations.
Tutorials
How-to guides
Explanations
Documentation boundaries¶
This site contains Flower Agent tutorials, guides, concepts, and operational
documentation. The generated flwr.agentapp Python API
reference remains
in the Flower framework documentation.
Design proposals and partially implemented paths aren’t documented as current capabilities. The documentation describes behavior available in the Flower framework and SuperGrid interfaces rather than planned product work.