logo

Introducing Kheish: An Open-Source Platform for Orchestrating Complex LLM Workflows

73 points graniet75
View on Reddit ↗

Hey everyone,

I’m excited to share Kheish, an open-source project I’ve been working on that aims to simplify the process of coordinating complex tasks using Large Language Models (LLMs).

What is Kheish?
Kheish helps you structure your LLM-based tasks through a multi-agent workflow. It assigns distinct roles—Proposer, Reviewer, Validator, and Formatter—to streamline the process from initial drafts to final, polished results. Instead of one big prompt and hope for the best, Kheish breaks down the journey into well-defined steps.

Key Features:

  • Multi-Agent Workflow: Each agent has a specific responsibility. The Proposer generates initial solutions, the Reviewer critiques and requests improvements, the Validator ensures everything meets the requirements, and the Formatter outputs the final result. This division leads to more controlled and reliable outcomes.
  • Extensible Modules: Kheish integrates seamlessly with external modules like fs (filesystem access), sh (shell commands), or vector_store (RAG support). Need to read a file, run a command, or perform a semantic search on large documents? Just call the relevant module. The results are cached and can be reused without re-running expensive operations.
  • Retrieval-Augmented Generation (RAG): Struggling with large codebases or documents? Kheish supports retrieval from vector stores, letting the LLM focus on only the most relevant context. No more flooding your prompt with excessive text.
  • Feedback Loops & Revisions: If the Reviewer or Validator isn’t satisfied, they can request revisions. This iterative approach helps you reach a better final solution without manually tweaking prompts over and over.

Who Might Benefit?

  • Developers building complex LLM-based tools and wanting more control over the generation process.
  • Projects requiring a structured approach to tackle large documents or code repositories.
  • Teams aiming for more predictable, testable LLM outputs.

Getting Started:
Check out the GitHub repo for documentation, examples, and instructions:
https://github.com/graniet/kheish

I’d love to hear your feedback, suggestions, or contributions. If you find Kheish helpful, feel free to open issues or submit PRs. Let’s collaborate to make LLM-based workflows more reliable, efficient, and user-friendly!

Thanks for reading, and happy experimenting with Kheish!