Prompts-GPT CLI Orchestration: Parallel, Pipeline & Eval Modes Explained

The Prompts-GPT CLI runs via npx prompts-gpt, which means no global installation is required. Node.js 18+ is the only prerequisite. Run npx prompts-gpt doctor --fix to verify your environment, check for common configuration issues, and auto-fix problems where possible.

For team setups, add prompts-gpt as a project dependency: npm install prompts-gpt. This ensures everyone on the team uses the same version. The CLI reads configuration from a local prompts-gpt.config file or environment variables for API keys and project settings.

Parallel mode (--mode parallel) runs multiple AI agents simultaneously against the same task. Each agent produces output independently, and the results are scored against configurable criteria. The best result is selected based on the scoring, giving teams a way to leverage multiple models or approaches without manual comparison.

Use parallel mode when the task has multiple valid approaches and you want to evaluate alternatives. For example: generating content briefs from the same visibility evidence using different AI models, or running a GEO content audit with different scoring configurations to see which produces the most actionable recommendations.

Pipeline mode (--mode pipeline) chains multiple steps sequentially with context passing between each step. Each step receives the output from the previous step as input, allowing teams to build complex multi-phase workflows from simple, composable steps.

A typical AI visibility pipeline might run: discover prompt gaps, generate content briefs from gaps, create draft content from briefs, evaluate drafts against GEO criteria, and publish approved content. Each step is independently configurable, testable, and replaceable without affecting the rest of the pipeline.

Eval mode (--mode eval) adds a self-evaluation scoring pass after each execution. Configure evaluation criteria such as correctness, completeness, risk, and relevance. Runs that score below a configurable threshold trigger automatic rollback or retry, ensuring that only quality output progresses through the workflow.

Eval mode is particularly useful for content production workflows where quality control is critical. A sweep that generates 20 content briefs can be evaluated automatically, with only briefs that pass quality criteria advancing to the review queue. This reduces manual review burden while maintaining quality standards.

The CLI integrates with CI/CD systems through standard exit codes and structured JSON output. Run npx prompts-gpt orchestrate in GitHub Actions, GitLab CI, or any CI system that supports Node.js. The --dry-run flag allows testing pipeline configurations without executing expensive AI operations.

For editor integration, the CLI works with Cursor, Claude Code, Copilot, ChatGPT, Gemini, Perplexity, Grok, and DeepSeek agents. The pipeline designer at prompts-gpt.com/orchestration provides a visual builder for multi-phase agent workflows with export to JSON, YAML, Bash, PowerShell, Docker, or GitHub Actions.

The CLI supports model aliases like fast, smart, cheap, and expensive that resolve to concrete AI models. This abstraction allows teams to switch between providers without changing pipeline configurations. Run npx prompts-gpt models to see available models, their cost tiers, and which alias each resolves to.

For cost management, combine model aliases with eval scoring. Run fast iterations during development with the cheap alias, then switch to the smart alias for production runs with eval gates ensuring quality. The --dry-run flag helps estimate costs before executing full pipelines.