agentops-toolkit 0.1.5

AgentOps CLI for standardized evaluation workflows

AgentOps Toolkit

AgentOps CLI for evaluation, observability, and operational workflows for Microsoft Foundry Agents and Models.

Overview

AgentOps Toolkit is a CLI built on Microsoft Foundry that standardizes evaluation and operational workflows for AI agents and models, helping teams run, monitor, and automate AgentOps processes.

The project enables:

• Consistent local and CI execution of agent evaluations
• Reusable evaluation policies through bundles
• Operational observability through tracing, monitoring, and run inspection
• Stable machine-readable outputs for automation
• Human-readable reports for PR reviews and quality gates

Operational capabilities include:

• Standardized evaluation workflows
• Run history and result inspection
• Tracing and observability
• Monitoring (dashboards and alerts)
• CI/CD automation
• Operational reporting and analysis

Core outputs:

• results.json (machine-readable)
• report.md (human-readable)

Exit code contract:

• 0 execution succeeded and all thresholds passed
• 2 execution succeeded but one or more thresholds failed
• 1 runtime or configuration error

Quickstart

1) Install

python -m venv .venv
# activate your venv in the current shell
python -m pip install -U pip
python -m pip install agentops-toolkit

2) Initialize and Configure

agentops init

This creates .agentops/ with starter bundles, datasets, and run configs for common scenarios (model quality, RAG, agent workflow, content safety).

Set your Foundry project endpoint:

export AZURE_AI_FOUNDRY_PROJECT_ENDPOINT="https://<resource>.services.ai.azure.com/api/projects/<project>"

Then edit .agentops/run.yaml to set your agent_id and model deployment name.

Authentication uses DefaultAzureCredential — run az login locally, or use service principal env vars in CI.

3) Run Evaluation

agentops eval run

Results are written to .agentops/results/latest/:

• results.json — machine-readable scores
• report.md — human-readable summary

To run a different scenario:

agentops eval run --config .agentops/run-rag.yaml

To regenerate the report from existing results:

agentops report generate

See docs/concepts.md for an overview of bundles, datasets, evaluators, backends, and the configuration model.

Commands

Command	Description	Status
agentops --version	Show installed version	✅
agentops init [--path DIR]	Scaffold project workspace, starter files, and coding agent skills	✅
agentops eval run [--config PATH]	Evaluate a dataset against a bundle	✅
agentops eval compare --runs ID1,ID2	Compare two past runs	✅
agentops report generate [--in FILE]	Regenerate report.md from results.json	✅
agentops workflow generate	Generate GitHub Actions workflow	✅
agentops skills install [--platform <p>]	Install coding agent skills (Copilot, Claude)	✅
agentops run list|show	List or inspect past runs	🚧
agentops bundle list|show	Browse bundle catalog	🚧
agentops dataset validate|describe	Dataset utilities	🚧
agentops trace init	Tracing setup	🚧
agentops monitor setup|show|configure	Monitoring operations	🚧

Planned commands return a friendly message indicating they are not yet implemented.

Documentation

Concepts and Architecture

• Concepts — bundles, datasets, evaluators, backends, configuration model
• How It Works — architecture, request flow, full schema reference
• Bundles — bundle authoring and evaluator configuration

Tutorials

• Model-direct evaluation
• Foundry agent evaluation
• RAG evaluation
• HTTP-deployed agent evaluation
• Conversational agent evaluation
• Agent workflow evaluation
• Baseline comparison

Operations

• CI/CD with GitHub Actions
• Copilot skills installation
• Release process
• Built-in evaluator reference

Contributing

See CONTRIBUTING.md for architecture rules, testing expectations, and contribution workflow.