Show HN: Pipelex – Declarative language for repeatable AI workflows (github.com)

Pipelex is an open-source declarative language and CLI for building repeatable, production-ready AI workflows. You can pip install it, run pipelex init, and create .plx files with a single build command (example: generate a CV-vs-job-offer pipeline). The language exposes typed Concepts (e.g., MatchAnalysis with strengths/gaps/areas_to_probe), modular Pipes (PipeSequence, PipeParallel, PipeExtract, PipeLLM), and explicit inputs/outputs (e.g., Question[5]). A provided CV→job-match example shows parallel OCR extraction of PDFs, an LLM-based structured match analysis (system_prompt + prompt), and a downstream LLM step that generates exactly five interview questions. Pipelines can be executed via CLI or asynchronously in Python (execute_pipeline). For practitioners this matters because Pipelex formalizes workflow structure, validation, and composability instead of ad-hoc monolithic prompts—improving reproducibility, observability and safer reuse. It supports BYO API keys (OpenAI, Anthropic, Google, Mistral, Bedrock, FAL) and local engines (Ollama, vLLM, llama.cpp) or custom endpoints. Features include IDE support (.plx extension), assistant-driven rule edits (natural-language pipeline tweaks), telemetry opt-ins, and an extensible extras install. The project is MIT-licensed, community-driven (Discord, GitHub), and geared toward turning multi-step AI tasks into auditable, typed pipelines ready for production.