agent-utilities

What it is and what it does

Agent Utilities is a framework for building AI agents on top of Pydantic-AI that bundles a knowledge graph, tool orchestration, memory, and skill management into a single harness. The zero-infrastructure default runs the knowledge graph in-process using the epistemic-graph Rust engine, with no external databases or services required to start. It exposes three consumption models: a Python library for standalone agents, an MCP server for integration with existing tools like Claude Code and Cursor, and a REST gateway for sharing a knowledge graph backend across multiple clients.

The package ships with configuration management via Pydantic Settings and environment variables, support for multiple deployment profiles (tiny for homelabs, single-node-prod, and enterprise), and a skill toolkit that auto-loads into agent tools. It handles secrets, multi-model configuration, and optional durable persistence to backends like Postgres or Neo4j. The core authority—the epistemic-graph engine—manages compute, caching, semantics, and persistence, while the gateway layer handles identity, action policies, and metrics.

Use it for:

• Build a standalone Python agent with built-in knowledge graph, memory, and tools without setting up external infrastructure.
• Integrate a knowledge graph and skill toolkit into Claude Code, Cursor, or other IDE-embedded AI tools via the MCP server.
• Deploy a shared knowledge graph backend across multiple agent clients or containers using the REST gateway.
• Set up an enterprise multi-host agent fleet with Vault, SSO, DNS, and observability using the enterprise profile.
• Ingest, query, and reason over domain knowledge using the in-process graph without a separate database.
• Evolve agent behavior by capturing learned skills and refined prompts back into a shared knowledge graph.

Worth the install?

AI-flagged interpretation of the facts on this page — verify before relying

A batteries-included harness for building Pydantic-AI agents with an in-process knowledge graph, orchestration, memory, and tools—deployable as a library, MCP server, or REST gateway.

Yes. The package is actively maintained, has low install friction, carries a permissive MIT license, and solves a real problem—bootstrapping Pydantic-AI agents with orchestration and knowledge graphs without external infrastructure. It's suitable for both rapid prototyping (tiny profile) and production deployments (enterprise profile). No known security vulnerabilities. The main gotcha is the Python 3.11+ requirement and the need to configure a model provider API key.

Install

agent-utilities on PyPI

pip install agent-utilities

uv add agent-utilities

poetry add agent-utilities

Installing agent-utilities

Before you install

Low install friction with a pure-Python wheel and 15 runtime dependencies. Active maintenance with a release 32 days ago. Requires Python 3.11 or later but supports current versions.

License in practice

MIT license is permissive—you can use, modify, and distribute this package freely in commercial and private projects with minimal restrictions.

Quickstart

pip install agent-utilities

from agent_utilities import create_agent

agent, toolsets = create_agent(name="assistant", skill_types=["universal", "graphs"])
print(agent.run_sync("What can you do?").output)

Requires Python 3.11 or later. A model provider API key (e.g., OPENAI_API_KEY) or local endpoint (vLLM/Ollama) must be configured via environment or .env file.

Verify before relying

• Whether the in-process knowledge graph persists across restarts or requires explicit serialization
• Performance characteristics and scalability limits of the embedded epistemic-graph engine
• Whether all 50+ connectors mentioned for enterprise profiles are included in the base package or require separate installation

Package facts

Evidence: agent_utilities-1.26.4-py3-none-any.whl

Tags

More Application Frameworks packages

Further reading