LazyCraft

LazyCraft is an AI Agent application development and management platform built on LazyLLM, designed to help developers quickly build and publish large model applications with low barriers and low cost.

Purpose

LazyCraft is aimed at non-developers and technical teams, and is used to quickly build a variety of AI applications, such as data insight analysts, literature review generators, code understanding assistants, etc., and supports publishing applications to a store or integrating them into business systems via API.

Main Capabilities

• Full-process closed loop: From application creation, debugging, and publishing to monitoring and Bad Case analysis, it fully covers the R&D chain and supports rapid prototype iteration all the way to production deployment.
• Flexible and pluggable architecture: Core modules are replaceable, compatible with multiple vector stores and RAG strategies, and support custom knowledge base orchestration, including customized offline parsing strategies and online recall strategies.
• Visual component canvas: Provides a visual component canvas to quickly build, debug, and publish AI applications.
• Built-in application templates: Built-in application templates are ready to use and help users get started quickly.
• Multi-model service integration: Integrates local and online model services from multiple model vendors, covering large language models, image-text understanding, text-to-image, speech-to-text, text-to-speech, embedding models, OCR, etc.
• Model management: Supports the complete process of inference → fine-tuning → evaluation, making it easy to switch models.
• Prompt management: Create and edit prompts, with built-in templates + AI-assisted writing.
• Knowledge base: Covers core components such as Reader, Rewrite, Retriever, and Rerank, covering the complete chain from document ingestion to retrieval; flexibly configure RAG strategies and multi-route recall; compatible with multiple file formats: PDF, Word, PPT, Excel, CSV, TXT, JSON, HTML, Markdown, LaTeX.
• Tools and MCP: Supports custom tools or API access, which can be called directly in the canvas.
• Datasets: Supports import of json, csv, and txt; built-in version management and template download; provides data cleaning, enhancement, and annotation capabilities.
• Permissions and security: Provides permission control and API Key management; integrates logs and auditing to meet enterprise security and compliance needs.
• Standardized interfaces: Provides standardized interfaces that can be seamlessly integrated into business systems.

Usage

• Deploy via Docker,You can use the official dev-latest image.
• HTTP can be used only when accessing via localhost/127.0.0.1; when accessing via a LAN IP or public IP, HTTPS must be used to protect data transmission security.
• To enable HTTPS access, perform the following steps:
  • This configuration provides an example configuration for a self-signed certificate. The browser will display a security warning; click "Continue to visit" to proceed.
  • If you need to use Let's Encrypt or a commercial CA certificate, apply for the certificate yourself, then modify the SSL certificate path configuration in docker/nginx/conf.d/default.conf.
• If the docker compose command cannot be found on mac, you can try installing docker via brew install --cask docker; or via brew install docker-compose, in which case replace docker compose up -d with docker-compose up -d.
• After docker starts, you can observe the startup status via the docker ps command and find the corresponding IP.
• After logging in, there will be some preset published applications. Before using these applications, make sure that the models they depend on have their Keys correctly configured. Sensenova models require applying for both ak and sk, and configuring them in the form ak:sk.
• Note: The following operations are all performed in a Linux environment.
• 1. Using the official dev-latest image: We will irregularly update and commit the latest changes to the dev-latest image, and use it as the latest version before the official release. When the image tag is latest, docker compose up -d will by default try to pull the latest latest from the remote. If you want to use latest but do not want to always pull updates, you can add the parameter --pull never.

Applicable scenarios

• Quickly build AI applications such as data insights, literature reviews, and code understanding.
• Teams that need low-code, component-based orchestration of AI applications.
• Scenarios that require custom RAG strategies and knowledge base orchestration.
• Scenarios that require integrated model fine-tuning, deployment, and evaluation.
• Scenarios that require integrating AI applications throughAPIs into business systems.

Limitations

• HTTPS must be used when accessing via a LAN IP or public IP.
• Self-signed certificates will trigger browser security warnings.
• Before using preset applications, ensure that the dependent models have their Keys correctly configured.
• Some operations need to be performed in a Linux environment.
• No other limitation information is currently provided.