cortexops 0.3.0

Reliability infrastructure for AI agents — evaluation, observability, and regression testing

CortexOps

Reliability infrastructure for AI agents.
Evaluate · Observe · Operate — for LangGraph, CrewAI, and AutoGen.

---

The problem

You deployed an agent. You have no idea if it regressed overnight.

No standard eval format. No failure traces. No CI gate before the next prompt change ships.
CortexOps fixes that.

---

Install

pip install cortexops

# With HTTP client (for pushing traces to hosted API):
pip install cortexops[http]

# With LLM judge support:
pip install cortexops[llm]

---

Quickstart

from cortexops import CortexTracer, EvalSuite

# Wrap your LangGraph app — zero refactor required
tracer = CortexTracer(project="payments-agent")
graph  = tracer.wrap(your_langgraph_app)

# Run evaluations against a golden dataset
results = EvalSuite.run(
    dataset="golden_v1.yaml",
    agent=graph,
)
print(results.summary())

---

Golden dataset (YAML)

version: 1
project: payments-agent

cases:
  - id: refund_lookup_01
    input: "What is the status of refund REF-8821?"
    expected_tool_calls: [lookup_refund]
    expected_output_contains: ["approved", "REF-8821"]
    max_latency_ms: 3000

  - id: open_ended_explanation_01
    input: "Why was my refund rejected?"
    judge: llm
    judge_criteria: >
      The response must explain the rejection reason clearly,
      be empathetic, and offer a concrete next step. No jargon.

---

CI gate

cortexops eval run \
  --dataset golden_v1.yaml \
  --fail-on "task_completion < 0.90"

Exits non-zero if the threshold is not met — blocks the PR.

---

Built-in metrics

Metric	What it checks
task_completion	Non-empty, non-error output with expected content
tool_accuracy	Expected tool calls were actually made
latency	Response within max_latency_ms budget
hallucination	Fabrication signals in output
llm_judge	GPT-4o scores against natural-language criteria

---

Links

• Docs: docs.cortexops.ai
• Repo: github.com/ashishodu2023/cortexops
• Issues: GitHub Issues