BalaganAgent¶
Chaos Engineering for AI Agents
Everyone demos agents. Nobody stress-tests them.
BalaganAgent is a reliability testing framework that stress-tests AI agents through controlled fault injection — because your agent will fail in production, and you should know how it handles it.
Why?¶
AI agents are entering production, but there's zero reliability discipline. Agents fail silently — tool calls time out, return garbage, or hallucinate. Context gets corrupted. Budgets get exhausted. Nobody knows until users complain.
BalaganAgent brings battle-tested chaos engineering principles (Chaos Monkey, Gremlin) to AI agents.
What you get¶
- :fontawesome-solid-bolt: Fault injectors — tool failures, delays, hallucinations, context corruption, budget exhaustion
- :fontawesome-solid-chart-line: Metrics — MTTR (Mean Time To Recovery) + SRE-grade reliability scoring
- :fontawesome-solid-file-lines: Reports — terminal / JSON / Markdown / HTML dashboards
- :fontawesome-solid-plug: Framework wrappers — CrewAI, AutoGen, LangChain, Claude Agent SDK
How it works¶
Install¶
60-second quickstart¶
from balaganagent import ChaosEngine, AgentWrapper
# 1. Wrap your agent
agent = MyAgent()
wrapper = AgentWrapper(agent)
wrapper.configure_chaos(chaos_level=0.5)
# 2. Calls now experience controlled failures
result = wrapper.call_tool("search", "test query")
# 3. Check metrics
metrics = wrapper.get_metrics()
print(f"MTTR: {metrics['mttr']}")
print(f"Reliability: {metrics['reliability_score']}")
Or use the CLI:
Example report output¶
Stress test results¶
Demo¶
Next steps¶
- How It Works — chaos levels, injectors, metrics in depth
- Integrations — CrewAI, AutoGen, LangChain, Claude SDK wrappers
- CLI Reference —
balaganagent run,stress,demo,init - Examples — real-world usage scripts
- FAQ — common questions
**"Hope is not a strategy. Test your agents."**
[GitHub](https://github.com/arielshad/balagan-agent){ .md-button } [PyPI](https://pypi.org/project/balagan-agent/){ .md-button .md-button--primary }