agno-agent

name: agno-agent description: | Build single Agno agents with tools, structured output, storage, memory, knowledge bases, guardrails, and human-in-the-loop confirmation. Trigger this skill when: importing agno.agent, creating an Agent instance, adding tools to an agent, configuring agent storage/memory, or asking "how do I build an agent with Agno?" license: Apache-2.0 metadata: version: "1.0.0" author: agno-team tags: ["agent", "agno", "ai", "tools", "structured-output"]

Build Agno Agents

Use agno.agent.Agent to create AI agents. Install with pip install agno.

Quick Start

from agno.agent import Agent
from agno.models.anthropic import Claude

agent = Agent(
    name="My Agent",
    model=Claude(id="claude-sonnet-4-5"),
    instructions=["Be concise.", "Use tables for data."],
    markdown=True,
)

agent.print_response("Hello!", stream=True)

Core Agent Parameters

Agent(
    name="Agent Name",
    model=Claude(id="claude-sonnet-4-5"),     # Required: model provider
    instructions="System prompt or list",      # str or list[str]
    tools=[ToolKit(), func],                   # Tools the agent can call
    markdown=True,                             # Format output as markdown
    show_tool_calls=True,                      # Display tool invocations
    add_datetime_to_context=True,              # Inject current datetime
    add_history_to_context=True,               # Include conversation history
    num_history_runs=5,                        # How many prior runs to include
    db=SqliteDb(db_file="agent.db"),           # Persistent storage backend
)

Agent with Tools

Give agents tools to interact with external data and services.

from agno.agent import Agent
from agno.models.anthropic import Claude
from agno.tools.yfinance import YFinanceTools

agent = Agent(
    name="Finance Agent",
    model=Claude(id="claude-sonnet-4-5"),
    tools=[YFinanceTools()],
    instructions="You are a data-driven financial analyst.",
    add_datetime_to_context=True,
    markdown=True,
)

agent.print_response("Give me a quick investment brief on NVIDIA", stream=True)

Structured Output

Use output_schema with a Pydantic model to get typed responses.

from typing import List, Optional
from pydantic import BaseModel, Field
from agno.agent import Agent
from agno.models.anthropic import Claude
from agno.tools.yfinance import YFinanceTools

class StockAnalysis(BaseModel):
    ticker: str = Field(..., description="Stock ticker symbol")
    current_price: float = Field(..., description="Current price in USD")
    summary: str = Field(..., description="One-line summary")
    key_risks: List[str] = Field(..., description="2-3 key risks")
    recommendation: str = Field(..., description="Buy, Hold, or Sell")

agent = Agent(
    name="Analyst",
    model=Claude(id="claude-sonnet-4-5"),
    tools=[YFinanceTools()],
    output_schema=StockAnalysis,
    markdown=True,
)

response = agent.run("Analyze NVIDIA")
analysis: StockAnalysis = response.content
print(f"Price: ${analysis.current_price:.2f}")
print(f"Recommendation: {analysis.recommendation}")

Typed Input and Output

Use input_schema + output_schema for end-to-end type safety.

from typing import List, Literal, Optional
from pydantic import BaseModel, Field
from agno.agent import Agent
from agno.models.anthropic import Claude

class AnalysisRequest(BaseModel):
    ticker: str = Field(..., description="Stock ticker symbol")
    analysis_type: Literal["quick", "deep"] = Field(default="quick")

class AnalysisResult(BaseModel):
    ticker: str
    summary: str
    recommendation: str

agent = Agent(
    name="Typed Agent",
    model=Claude(id="claude-sonnet-4-5"),
    input_schema=AnalysisRequest,
    output_schema=AnalysisResult,
)

# Pass input as dict or Pydantic model
response = agent.run(input={"ticker": "NVDA", "analysis_type": "deep"})
result: AnalysisResult = response.content

Storage (Persistent Conversations)

Add db= to persist conversation history across runs. Use session_id to maintain conversation threads.

from agno.agent import Agent
from agno.models.anthropic import Claude
from agno.db.sqlite import SqliteDb

agent = Agent(
    name="Persistent Agent",
    model=Claude(id="claude-sonnet-4-5"),
    db=SqliteDb(db_file="tmp/agents.db"),
    add_history_to_context=True,
    num_history_runs=5,
)

# Same session_id = continuous conversation
agent.print_response("Hello!", session_id="my-session", stream=True)
agent.print_response("What did I just say?", session_id="my-session", stream=True)

Memory (User Preferences)

Memory persists user-level facts across all sessions. Different from storage (conversation history).

from agno.agent import Agent
from agno.models.anthropic import Claude
from agno.db.sqlite import SqliteDb
from agno.memory import MemoryManager

db = SqliteDb(db_file="tmp/agents.db")

memory_manager = MemoryManager(
    model=Claude(id="claude-sonnet-4-5"),
    db=db,
    additional_instructions="Capture user preferences and goals.",
)

agent = Agent(
    name="Memory Agent",
    model=Claude(id="claude-sonnet-4-5"),
    db=db,
    memory_manager=memory_manager,
    enable_agentic_memory=True,    # Agent decides when to store/recall
    # OR: update_memory_on_run=True  # Always run memory manager (guaranteed but slower)
    add_history_to_context=True,
    num_history_runs=5,
)

agent.print_response("I prefer tech stocks", user_id="user@example.com", stream=True)
memories = agent.get_user_memories(user_id="user@example.com")

State Management

Use session_state for structured data the agent actively manages (watchlists, counters, flags).

from agno.agent import Agent
from agno.models.anthropic import Claude
from agno.db.sqlite import SqliteDb
from agno.run import RunContext

def add_item(run_context: RunContext, item: str) -> str:
    """Add an item to the list."""
    items = run_context.session_state.get("items", [])
    items.append(item)
    run_context.session_state["items"] = items
    return f"Added {item}. Total: {len(items)}"

agent = Agent(
    name="Stateful Agent",
    model=Claude(id="claude-sonnet-4-5"),
    tools=[add_item],
    session_state={"items": []},
    add_session_state_to_context=True,
    db=SqliteDb(db_file="tmp/agents.db"),
    instructions="Current items: {items}",  # State injected into instructions
)

Access state: agent.get_session_state() or response.session_state.

Knowledge Base

Give agents searchable document collections using vector databases.

from agno.agent import Agent
from agno.models.anthropic import Claude
from agno.db.sqlite import SqliteDb
from agno.knowledge.knowledge import Knowledge
from agno.knowledge.embedder.google import GeminiEmbedder
from agno.vectordb.chroma import ChromaDb
from agno.vectordb.search import SearchType

db = SqliteDb(db_file="tmp/agents.db")

knowledge = Knowledge(
    name="My Docs",
    vector_db=ChromaDb(
        name="docs",
        collection="docs",
        path="tmp/chromadb",
        persistent_client=True,
        search_type=SearchType.hybrid,
        embedder=GeminiEmbedder(id="gemini-embedding-001"),
    ),
    max_results=5,
    contents_db=db,
)

agent = Agent(
    name="Knowledge Agent",
    model=Claude(id="claude-sonnet-4-5"),
    knowledge=knowledge,
    search_knowledge=True,  # Agent searches knowledge automatically
    db=db,
)

# Load documents
knowledge.insert(name="Docs", url="https://docs.agno.com/introduction.md")
agent.print_response("What is Agno?", stream=True)

Guardrails

Validate input before processing using pre_hooks.

from agno.agent import Agent
from agno.models.anthropic import Claude
from agno.guardrails import PIIDetectionGuardrail, PromptInjectionGuardrail
from agno.guardrails.base import BaseGuardrail
from agno.exceptions import InputCheckError
from agno.run.agent import RunInput

class SpamGuardrail(BaseGuardrail):
    def check(self, run_input):
        content = run_input.input_content_string()
        if content.count("!") > 5:
            raise InputCheckError("Input appears to be spam")

    async def async_check(self, run_input):
        self.check(run_input)

agent = Agent(
    name="Safe Agent",
    model=Claude(id="claude-sonnet-4-5"),
    pre_hooks=[
        PIIDetectionGuardrail(),
        PromptInjectionGuardrail(),
        SpamGuardrail(),
    ],
)

Human-in-the-Loop

Require user confirmation before executing sensitive tools.

from agno.agent import Agent
from agno.models.anthropic import Claude
from agno.tools import tool

@tool(requires_confirmation=True)
def delete_record(record_id: str) -> str:
    """Delete a record. Requires user confirmation."""
    return f"Deleted record {record_id}"

agent = Agent(
    name="Careful Agent",
    model=Claude(id="claude-sonnet-4-5"),
    tools=[delete_record],
)

response = agent.run("Delete record ABC-123")

# Check for pending confirmations
if response.active_requirements:
    for req in response.active_requirements:
        if req.needs_confirmation:
            req.confirm()  # or req.reject()

    # Resume execution
    response = agent.continue_run(
        run_id=response.run_id,
        requirements=response.requirements,
    )

Model Providers

Swap models by changing the import and model ID:

from agno.models.anthropic import Claude
model = Claude(id="claude-sonnet-4-5")

from agno.models.openai import OpenAIChat
model = OpenAIChat(id="gpt-4o")

from agno.models.google import Gemini
model = Gemini(id="gemini-2.0-flash")

from agno.models.ollama import Ollama
model = Ollama(id="llama3.3")

Anti-Patterns

Don't forget add_history_to_context=True if you want conversational agents that remember prior messages
Don't skip db= if you want sessions to persist across script restarts
Don't hardcode model IDs without showing how to swap them — use a config pattern
Don't use OpenAIChat when you want the newer Responses API — use OpenAIResponses instead
Don't use blocking calls in async contexts — use agent.arun() and agent.aprint_response() for async
Don't forget to install extras: pip install "agno[anthropic]" for Claude, pip install "agno[openai]" for OpenAI