Agents
Steamship is made for building, scaling, and managing Agents in the cloud.
A full tutorial for getting started with Agents is available here:
Agent Guidebook (opens in a new tab)
If you’re new to building Agents, we highly recommend following the above tutorial.
The following are pointers to the key Agent classes for getting started:
Agent
The basic interface for Agents is:
steamship.agents.schema.agent.Agent
Implementations of this class make decisions about what Action will be taken next in an Agent workflow.
AgentContext
The Agent Context contains data about the context in which the Agent is running, including its ChatHistory.
Tool
A Tool is something that an Agent may make use of in service of achieving its goal. To support multi-modal data, Tools in Steamship take an AgentContext as input and produce Blocks as output.
Many tools are available pre-packaged within the SDK:
steamship.agents.tools.search.search.SearchToolsteamship.agents.tools.speech_generation.generate_speech.GenerateSpeechTool(via ElevenLabs)steamship.agents.tools.image_generation.dalle.DalleToolsteamship.agents.tools.image_generation.stable_diffusion.StableDiffusionTool- … and more in the
steamship.agents.toolspackage.
AgentService
steamship.agents.service.agent_service.AgentService
The AgentService class provides a convenient way to deploy an Agent as a Steamship Package.
All steamship.agents.service.agent_service.AgentService instances contain a prompt(self, prompt: str, **kwargs) -> List[Block] method from thier base class.
This method is the core chat loop: it accepts an inbound str in the form of a user message, and it produces a list of multimodal Block objects that contain the response.