Getting started
A little preparation. A clearer first conversation.
Start small: one concrete question, relevant AI roles, and the information they need to help.
1. Check Windows access
CrewParley’s maintained packaging target is Windows x64. Public distribution is still pending, so this site does not provide an installer. See access information for the current state.
Use an approved distribution when one is published. Check its version, source, and any provided checksum or signing information. Never disable antivirus or bypass a security warning to install the software.
2. Prepare your AI tools
Supported provider command-line tools must be installed and authenticated separately. You may need a provider account, subscription, or usage credits. Review the provider’s current setup instructions and policies.
The preview has supplied-context drafting evidence for Codex and Claude Code. Complete installed workflows still require release validation. Provider and connection details explain the boundaries.
3. Choose relevant AI roles
A role provides instructions and a focus for an AI agent. Choose roles suited to the question—for example, a planner to structure an approach and a reviewer to challenge assumptions.
Named agents and AI teams are not additional human collaborators. A profile does not connect email, calendars, or business systems.
4. Supply the context that matters
State the desired outcome, constraints, and information you want the roles to consider. Only include information you are entitled to share with the selected provider.
Choose a conversation to explore a question, a task workflow to coordinate work, or a debate to compare perspectives. Shared context is bounded; avoid assuming every prior discussion or file is included.
5. Review outputs and decide
Read the result and its context. Verify important facts and inspect any actual changes. An agent reporting completion does not prove correctness.
Decide whether to revise the question, ask another role to contribute, or take the next step yourself. External actions depend on the actual tool permissions; review controls are not universally enforced across all workflows.