Wolfpack AI

Wolfpack AI is a Python framework for building production-ready artificial intelligence agents. It combines agent execution, tool calling, retrieval-augmented generation (RAG), memory, human approval flows, guardrails, and OpenTelemetry observability in a single developer-focused library.

Every agent run can emit structured traces for model calls, tool executions, token usage, latency, and cost. The optional Wolfpack Agent Management Platform (AMP) receives those traces and provides an operational interface for monitoring, governance, evaluation, and agent lifecycle management.

Install

Wolfpack AI requires Python 3.10 or later.

pip install wolfpackai

The published package is named wolfpackai; Python imports remain under the stable wolfpack module.

Install optional integrations as needed:

pip install "wolfpackai[qdrant]"    # Qdrant vector database
pip install "wolfpackai[pgvector]"  # PostgreSQL vector search
pip install "wolfpackai[knowledge]" # Document readers
pip install "wolfpackai[mcp]"       # Model Context Protocol client

Quick Start

Configure a supported model provider, then create an agent and attach typed Python functions as tools.

export OPENAI_API_KEY="..."

from wolfpack import Agent, get_model_from_env, tool


@tool
def get_weather(city: str) -> str:
    """Return the current weather for a city.

    Args:
        city: City name.
    """
    return f"{city}: 22 C, cloudy."


agent = Agent(
    name="Weather assistant",
    model=get_model_from_env(),
    tools=[get_weather],
)

result = agent.run("What is the weather in Lisbon?")
print(result.content)

Capabilities

• Agent loop with synchronous execution, asynchronous execution, streaming, events, and typed tool calling.
• Model adapters for OpenAI, Anthropic, Google Gemini, Ollama, Groq, and OpenAI-compatible providers.
• Knowledge retrieval with chunking, embeddings, and memory, Qdrant, or PostgreSQL vector stores.
• Persistent session memory with in-memory and SQLite stores.
• Guardrails for personally identifiable information, prompt injection, and tool allowlists.
• Durable human-in-the-loop approvals, including pause and resume workflows.
• Pydantic-based structured output with validation and retry feedback.
• Deterministic workflows with directed acyclic graph execution, conditions, and retries.
• Multi-agent teams with delegation, routing, broadcast, and task modes.
• Model Context Protocol client support for external tool servers.
• Evaluation runners, score publishing, scheduled tasks, and channel adapters for Telegram, Slack, Discord, and web chat.

Observability

Wolfpack AI emits OpenTelemetry spans using the generative artificial intelligence semantic conventions. Set an Agent Management Platform endpoint and project Application Programming Interface key to send runs to the control plane:

export WOLFPACK_AMP_URL="http://localhost:8000"
export WOLFPACK_AMP_API_KEY="pk-...:secret"

The Agent Management Platform stores traces and provides dashboards, trace exploration, session views, approvals, guardrail settings, evaluation scores, privacy controls, schedules, runtime mesh monitoring, and channel management.

Development

Clone the repository and install the development dependencies with uv:

git clone https://github.com/wolfpackaiagents/wolfpackai.git
cd wolfpackai/framework
uv sync --extra test --group dev
pytest

Examples are organized by capability in examples/, including basic agents, tools, knowledge retrieval, observability, human approval, workflows, teams,

License

Wolfpack AI is released under the MIT License.