Kheish is an open-source multi-agent coordination platform that allows for flexible configuration of multiple agents to solve complex tasks.
Kheish is an open-source, multi-agent orchestration platform based on Large Language Models (LLMs). It coordinates the various steps of complex tasks, such as proposal generation, review, validation, and..., using multiple specialized roles (agents) and flexible workflows.
What is Kheish?
Kheish is an open-source, multi-agent orchestration platform based on Large Language Models (LLMs). It coordinates the various steps of complex tasks, such as proposal generation, review, validation, and formatting, using multiple specialized roles (agents) and flexible workflows to produce high-quality results. The platform seamlessly integrates external modules, such as file system access, shell commands, and vector storage, enriching the context and handling large codebases or documents. With its modularity, scalability, feedback loops, and chat-like prompts, Kheish supports customization and expansion to new tasks or domains.
Kheish's main functions
- Multi-agent workflowDefine multiple roles (proposer, reviewer, verifier, formatter) and execute them in the order of the YAML-based workflow configuration to handle complex tasks.
- Module integrationIt integrates multiple modules, such as file system access (fs), shell commands (sh), and vector storage (vector_store), to expand functionality and processing capabilities.
- Chat-style promptsInteraction with the LLM uses a dialogic structure, including system, user, and assistant, to maintain context and clarify instructions.
- Feedback and revision cycleThe workflow supports iterative feedback, allowing reviewers or verifiers to request revisions if they find issues, until the solution meets the standards.
- RAG and EmbeddedIntegrated vector storage allows for the processing of large documents, enabling the model to retrieve relevant fragments without overloading the prompts.
- Modularity and scalabilitySupports adding new modules or agents without breaking the existing structure, and encourages customization and expansion to new tasks or domains.
Kheish's technical principles
- task managerDesign a task manager that is responsible for loading task configuration (YAML), handling context, running workflows, and integrating modules as needed.
- Context processingThe Task Manager reads the initial file or text, prepares the TaskContext, and provides context information for subsequent agent workflows.
- Workflow ExecutionThe agent executes tasks according to the defined workflow sequence, including generating solutions, checking proposal quality, ensuring final correctness, and outputting verification results.
- Module request callIntelligent agents can issue MODULE_REQUEST calls to obtain other data, read files, or perform semantic searches, etc.
- Result CacheTo avoid redundant operations, Kheish caches the results of the proxy request module.
- Integration with RAGIt integrates with the Retrieval Enhanced Generation (RAG) model to easily process large documents and retrieve relevant fragments via vector storage.
Kheish's project address
- GitHub repository:https://github.com/graniet/kheish
Kheish application scenarios
- Code auditUsed for automated code review, it identifies security vulnerabilities and quality issues in code through intelligent agent workflows and proposes improvement suggestions.
- Legal document analysisIn the legal field, this involves processing and analyzing large amounts of legal documents, extracting key information, and assisting in legal research and case preparation.
- Customer service automationAs a chatbot, it handles customer inquiries and support requests, providing instant automated responses.
- Content Creation AssistantIt assists content creators in generating and editing articles, blogs, and other marketing materials, improving the quality and appeal of their content.
- Personalized education platformProvide personalized learning materials and exercises based on students' learning progress to support online education and remote learning.