AgentGraph visualization

AgentGraph walks an agent’s configuration and produces a graph of its wiring. Workers, tools, sub-architectures, retrievers. Useful for:

• Documentation. Auto-render the team layout into your README or design doc.
• Debugging. See at a glance whether a worker is connected where you think it is.
• Onboarding. A new engineer reads the diagram before reading the code.

Quick render

from loomflow import Agent
from loomflow.graph import build_graph, write_graph
from loomflow.team import Team
 
team = Team.supervisor(
    workers={"researcher": researcher, "writer": writer, "reviewer": reviewer},
    instructions="manage the pipeline",
    model="claude-opus-4-7",
)
 
graph = build_graph(team)
print(graph.to_mermaid())

Output (Mermaid):

Write to disk

write_graph(team, path="agent.mmd",  format="mermaid")    # Mermaid syntax
write_graph(team, path="agent.dot",  format="graphviz")   # Graphviz DOT
write_graph(team, path="agent.json", format="json")       # raw structure

The Mermaid file renders directly on GitHub / GitLab / Notion. The Graphviz file becomes a PNG / SVG via dot agent.dot -Tsvg -o agent.svg.

What gets included

AgentGraph walks:

• The top-level architecture and its declared workers.
• Each worker’s own architecture (recursively. Nested Supervisors, Reflexion-of-Supervisor, etc.).
• Tools registered at each level (when include_tools=True).
• MCP servers and their tool surface (when include_mcp=True).

graph = build_graph(team, include_tools=True, include_mcp=True)

Why not OTel + Tempo / Jaeger?

OTel traces show per-run dynamics. What happened during this specific agent.run(). AgentGraph shows static wiring. The shape of your agent regardless of input. Both are useful; they answer different questions.