Security advisory — Malicious litellm versions 1.82.7 and 1.82.8 were removed from PyPI (potential API key exfiltration). Uninstall them, rotate exposed credentials, and upgrade to a safe release (e.g. 1.82.9+). Run pip show litellm to verify. PyPI · README

Building Your First Agent

Step-by-step guide to building a research agent.

Building Your First Agent

Build a research agent that takes a topic, calls tools, and returns a structured report. This is the SDK path: Agent + Task + AgentExecutor. For Near chat, start agx serve instead — see Quickstart.

Diagram: Task into AgentExecutor.run; tool_calls loop until a text report.

Step 1: Set up

bash
1pip install agenticx
2export OPENAI_API_KEY="your-key"

Step 2: Define tools

python
1from agenticx.tools import tool
2import httpx
3

4@tool
5def search_web(query: str) -> str:
6    """Search the web for information about a topic.
7

8    Args:
9        query: The search query
10    """
11    response = httpx.get(
12        "https://api.search.com/search",
13        params={"q": query, "key": "your-api-key"},
14        timeout=30,
15    )
16    return response.text
17

18@tool
19def fetch_page(url: str) -> str:
20    """Fetch the first 5000 characters of a web page.
21

22    Args:
23        url: The URL to fetch
24    """
25    response = httpx.get(url, follow_redirects=True, timeout=30)
26    return response.text[:5000]

Replace the search URL with a real API. @tool wraps a function as a FunctionTool / BaseTool.

Step 3: Define the agent

python
1from agenticx import Agent
2

3research_agent = Agent(
4    id="research-agent",
5    name="Research Assistant",
6    role="Senior Research Analyst",
7    goal=(
8        "Conduct thorough research on any given topic. "
9        "Find authoritative sources, synthesize information, "
10        "and produce clear, well-structured reports."
11    ),
12    backstory=(
13        "You are an expert researcher with a background in "
14        "information synthesis and critical analysis."
15    ),
16    organization_id="my-research-org",
17    max_iterations=15,
18    tools=[search_web, fetch_page],
19)

Step 4: Run a task

python
1from agenticx import Task, AgentExecutor
2from agenticx.llms import OpenAIProvider
3

4task = Task(
5    description="Research: Multi-agent AI systems and software development",
6    expected_output=(
7        "A structured research report with:\n"
8        "1. Executive summary\n"
9        "2. Key findings\n"
10        "3. Detailed analysis\n"
11        "4. Sources and references"
12    ),
13)
14

15executor = AgentExecutor(
16    llm_provider=OpenAIProvider(model="gpt-4o"),
17    tools=[search_web, fetch_page],
18)
19result = executor.run(agent=research_agent, task=task)
20print(result)

run returns a dict with the final output and metadata, not a bare string.

Next