Skip to content

IBM watsonx Orchestrate

Important: Hosted evaluation commands require authorized IBM credentials and a known environment. Historical M6.3/M6.4 configurations are not equivalent to the later consolidated 29-tool server and 21-tool Avery attachment configuration.

# Isolated IBM ADK environment (do not alter the backend Python environment)
python -m venv .venv-adk
.\.venv-adk\Scripts\Activate.ps1
python -m pip install --upgrade ibm-watsonx-orchestrate
orchestrate --version
orchestrate --help
orchestrate env --help
orchestrate env list

# Select an existing authorized environment
$EnvironmentName = 'YOUR_AUTHORIZED_ENVIRONMENT'
orchestrate env activate $EnvironmentName
orchestrate agents list
orchestrate toolkits list
orchestrate connections list
orchestrate evaluations --help

Project-owned evaluation runners (record the environment and output directory):

pwsh ./deploy/ibm/orchestrate/run-m64-evaluation.ps1 -EnvironmentName $EnvironmentName
pwsh ./deploy/ibm/orchestrate/run-m63-evaluation.ps1 -EnvironmentName $EnvironmentName
pwsh ./deploy/ibm/orchestrate/run-m63-regression.ps1 -EnvironmentName $EnvironmentName

M6.4 historical runner uses 13 scenario JSON files; that number does not imply 13/13 passed. M6.3 historical evaluation uses 10 direct cases and 6 focused regression cases. Inspect each newly written summary_metrics.csv, raw traces and run identity. The November 2026 evidence is versioned and must not be merged across configurations.

A general current IBM ADK pattern is orchestrate evaluations evaluate --help and the IBM evaluation guide. Consult that command's output for your installed ADK version before supplying flags. Do not rerun original configure-m64.ps1 against the accepted later Avery deployment; it may replace its earlier candidate toolkit. Project IBM CLI runbook.


Developer Guide: Overview · ← Supabase · DeepEval →