Agloom

Agloom is a production-grade Python agent harness built on LangChain and LangGraph. It keeps straightforward requests on lightweight DIRECT execution and uses one of eight composable execution topologies when a task benefits from tools, planning, workers, reflection, or explicit composition.

Agloom provides eight execution topologies plus a direct execution mode. DIRECT is not a topology.

What developers get

• one construction API: create_agent();
• automatic, constrained, or explicit execution selection;
• local LangChain tools, application capabilities, and MCP capabilities in one policy-aware catalog;
• bounded recursive execution, streaming, checkpointing, and human approval;
• Prometheus, Grafana, Langfuse, and langgraph-xai observability;
• structured output through xstructured;
• a portable Agent Skill for coding-agent integrations.

Start here

• Understand why Agloom exists
• Install Agloom
• Build your first agent
• Browse topology concepts
• Review examples and verified results
• Learn how to choose an execution mode
• Explore the complete create_agent() reference

Design

The public API stays small:

from agloom import create_agent

agent = create_agent(model=model, tools=tools)
answer = agent.invoke("Use the configured tools only when the request needs them.")
print(answer.content)

An unconstrained agent uses its LLM to analyze requests against the configured capability inventory, then chooses DIRECT or a topology. Missing required capabilities fail explicitly. An explicit single pattern bypasses selection; an allow-list constrains automatic selection. See choosing an execution mode.