"""
Example: Per-Hook Background Control with AgentAsJudgeEval in AgentOS
This example demonstrates fine-grained control over which hooks run in background:
- Set eval.run_in_background = True for eval instances
- AgentAsJudgeEval evaluates output quality based on custom criteria
"""
from agno.agent import Agent
from agno.db.sqlite import AsyncSqliteDb
from agno.eval.agent_as_judge import AgentAsJudgeEval
from agno.models.openai import OpenAIChat
from agno.os import AgentOS
# ---------------------------------------------------------------------------
# Create Example
# ---------------------------------------------------------------------------
# Setup database
db = AsyncSqliteDb(db_file="tmp/agent_as_judge_evals.db")
# AgentAsJudgeEval for completeness - runs synchronously (blocks response)
completeness_eval = AgentAsJudgeEval(
db=db,
name="Completeness Check",
model=OpenAIChat(id="gpt-5.2"),
criteria="Response should be thorough, complete, and address all aspects of the question",
print_results=True,
print_summary=True,
telemetry=True,
)
# completeness_eval.run_in_background = False (default - blocks)
# AgentAsJudgeEval for quality - runs in background (non-blocking)
quality_eval = AgentAsJudgeEval(
db=db,
name="Quality Assessment",
model=OpenAIChat(id="gpt-5.2"),
criteria="Response should be well-structured, concise, and professional",
scoring_strategy="numeric",
threshold=8,
additional_guidelines=[
"Check if response is easy to understand",
"Verify response is not overly verbose",
],
print_results=True,
print_summary=True,
run_in_background=True, # Run this eval as a background task
)
agent = Agent(
id="geography-agent",
name="GeographyAgent",
model=OpenAIChat(id="gpt-5.2"),
instructions="You are a helpful geography assistant. Provide accurate and concise answers.",
db=db,
post_hooks=[
completeness_eval, # run_in_background=False - runs first, blocks
quality_eval, # run_in_background=True - runs after response
],
markdown=True,
telemetry=False,
)
# Create AgentOS
agent_os = AgentOS(agents=[agent])
app = agent_os.get_app()
# Flow:
# 1. Agent processes request
# 2. Sync hooks run (completeness_eval)
# 3. Response sent to user
# 4. Background hooks run (quality_eval)
# Test with:
# curl -X POST http://localhost:7777/agents/geography-agent/runs \
# -F "message=What is the capital of France?" -F "stream=false"
# ---------------------------------------------------------------------------
# Run Example
# ---------------------------------------------------------------------------
if __name__ == "__main__":
agent_os.serve(app="background_evals_example:app", port=7777, reload=True)
