dralius97/covalve — reverse-engineered prompt

Build me a small Python library called covalve that lets developers run AI chat workflows in a predictable state machine instead of letting the model decide everything.

It should expose a simple pipeline function, a default conversational pipeline, and clear interfaces people can implement for their own LLM, memory storage, cache, tools, logging, and guardrails. The user should be able to pass a query and optional session id, then get back a structured response.

Include a built in schema driven flow for loading chat history, analyzing intent, mapping tools, executing tools by priority, retrying on errors, generating a final answer, and saving the conversation. Make tool routing configurable with a JSON style tools schema.

Also support custom pipeline states, replacing built in handlers, and lightweight hooks so people can log or block specific state transitions without changing the core executor.

Keep it Python 3.10 plus, async friendly, typed, and easy to install. Add examples and enough README docs so another developer can wire it to OpenAI, Gemini, Ollama, or their own infrastructure.