Skip to main content
What this builds. A functional agent that takes a typed topic input, produces a typed brief output, and ships with a Google Search tool. You’ll end up with. A saved agent whose inputs, outputs, tools, and agent_type you can verify locally before running it. Demonstrates creating a functional agent with typed IoConfig inputs/outputs and tool references in instructions using {{}} placeholders (v2 feature, shown here for illustration). Note: {{}} template validation is a v2 feature. In v1, instructions are plain strings and {{}} references are passed through as-is. 
from vectorshift.agent import Agent, AgentType, IoConfig, LlmInfo
from vectorshift.agent.tool import ToolInput, ToolInputType
from vectorshift.agent.tools import ExaAiTool, GoogleSearchTool

# Create a tool instance directly
search = GoogleSearchTool(
    query=ToolInput(type=ToolInputType.DYNAMIC, description=None),
    num_results=10,
)

exa_search = ExaAiTool(
    tool_name="exa_ai_search",
    query=ToolInput(type=ToolInputType.DYNAMIC, description="Search query"),
)

# Create a functional agent with typed inputs and outputs
agent = Agent.new(
    name="Topic briefing",
    type=AgentType.FUNCTIONAL,
    llm_info=LlmInfo(provider="openai", model_id="gpt-4o"),
    tools=[search],
    instructions=(
        "For the given topic, search the web if you need "
        "fresh facts, then write a short brief."
    ),
    inputs={
        "topic": IoConfig(io_type="string", description="Subject to brief"),
    },
    outputs={
        "brief": IoConfig(io_type="string", description="Markdown brief"),
    },
)

print(f"Created functional agent: {agent.name} (id={agent.id})")
print(f"  Inputs:  {list(agent.inputs.keys())}")
print(f"  Outputs: {list(agent.outputs.keys())}")
print(f"  Tools:   {[t.name for t in agent.tools]}")
print(f"  Type:    {agent.agent_type}")

# Clean up
agent.delete()
print("Deleted agent.")

​
Expected output

Created functional agent: Topic briefing (id=...)
  Inputs:  ['topic']
  Outputs: ['brief']
  Tools:   ['google_search']
  Type:    AgentType.FUNCTIONAL
Deleted agent.
agent.id is server-assigned, so it changes every run; the rest of the lines reflect exactly what you declared.

​
See also

Functional agent run

Take this typed agent and execute it with agent.run(inputs=...).

Conversational agent with tools

The conversational counterpart — tools managed via the fluent API.

Agent reference

Full method surface.