DOCSEARCHbyAlgolia
Submit Docs

CSGHub Documentation

website·Indexed 32 days ago

https://opencsg.com/docs

Documentation for CSGHub, a distributed cloud-native AI hosting platform. Includes installation, configuration, and upgrade guides for Kubernetes and Docker deployments, covering versions v2.2.0 and v2.4.0, as well as details on core components like the Runner and CSGShip.

Tokens
1.2M
Snippets
4.1K
Records
10.1K
Agent score
49%

What's inside CSGHub

  1. Overview of CSGHub Model Asset Management Platform

    CSGHub is an open-source large model digital asset management platform designed for enterprise AI construction, domestic computing power, and open-source model ecosystem management. Unlike cloud-dependent alternatives, it supports complete air-gapped (offline) deployment in corporate intranets and is compatible with multiple chip architectures and hardware environments.

    Key management capabilities include:

    • Asset Types: Models (weights, config, code), Datasets (structured/unstructured), PromptSets, Agent applications (Spaces), and MCP (Servers + Tools).
    • Management Features: Version control, metadata traceability, hierarchical permissions, and security auditing.
    • Distribution: Supports S3, Git LFS, Web UI, REST API, and SDKs.
    • Execution: Supports multi-GPU fine-tuning, high-performance inference, and data processing.

  2. Overview of XNet for AI/ML Storage

    XNet is a high-performance storage backend designed specifically for modern AI workflows involving ultra-large models and massive datasets. Unlike Git LFS, which operates on whole files, XNet supports fine-grained data access, intelligent caching, and parallel transmission. Key capabilities include:

    • Intelligent Deduplication: Files are split into blocks; only unique blocks are stored across versions and projects, reducing storage and increasing upload speeds.
    • Incremental Updates: Only changed data blocks are uploaded, reducing update times to seconds for iterative model/dataset changes.
    • Parallel Downloads: Uses a streaming block-download mechanism to maximize bandwidth, ideal for training clusters and inference services.
    • Data Integrity: Every block is uniquely identified via cryptographic hashes, ensuring end-to-end integrity and efficient synchronization across regions.

  3. Overview of the StarChain LLM Application Platform

    StarChain is an open-source, low-code platform designed for building, testing, hosting, and sharing multimodal LLM applications. It supports various mainstream LLM chain frameworks and integrates with private, open-source, and commercial LLM backends. The platform provides tools for orchestration, testing, and application sharing.

  5. Overview of CSGHUB Dataflow

    v1.16.0
    CSGHUB Dataflow is a data flow management and labeling subsystem within the CSGHub platform. It is used for processing model training data, managing labeling tasks, and handling data preprocessing and distribution. It can be deployed on Kubernetes via Helm Chart and includes dependencies such as Label Studio, Redis, PostgreSQL, and MongoDB.

  6. Overview of AgenticHub capabilities

    AgenticHub is a module within the CSGHub platform used to transform conversational models into task-oriented, reusable, and governable AI agents. It integrates visual orchestration, model invocation, API integration, tool/data access, and process orchestration. Key capabilities include:

    • Agent Creation: Use prompts and conversational guidance to define executable plans.
    • Model Selection: Switch underlying LLMs based on task type to balance performance, cost, and latency.
    • Skills: Use built-in skills or upload custom skills as pluggable components for team sharing.
    • Connectivity: Integrate with collaboration channels (Feishu, WeChat) via OpenClaw and extend tools/data access using the Model Context Protocol (MCP).
    • Long-term Memory: Persist user preferences, agreements, and conclusions across multiple sessions to reduce repetitive input.
    • Automation: Schedule recurring tasks using the 'Scheduled Tasks' feature to execute preset prompts.
    • RAG & Resources: Combine private data, knowledge bases, and Retrieval-Augmented Generation (RAG) for grounded responses.

  7. Overview of Text Generation Tasks

    Text Generation is a core LLM inference task that automatically generates coherent text based on an input prompt. Common application scenarios include:

    • Content Creation: Articles, stories, poetry, and advertising copy.
    • Dialogue Systems: Intelligent customer service and chatbots.
    • Code Generation: Automatically generating code from requirement descriptions.
    • Text Summarization: Extracting core information from long documents.
    • Machine Translation: Translating text between different languages.

  8. Overview of CSGHub Enterprise AI Platform

    CSGHub is a comprehensive one-stop management platform for enterprise-level AI research and asset control. It provides a full-link closed-loop capability covering asset management, compute resource scheduling, model development, and service commercialization. It supports various user roles including Algorithm Engineers, Business Developers, System Administrators, Agents/Third-party applications, and Data Labelers.

  9. Overview of the StarChain LLM Application Platform

    StarChain is an open-source, low-code platform designed for the construction, testing, hosting, and sharing of multimodal Large Language Model (LLM) applications. It supports integration with various mainstream LLM chain frameworks and allows the use of private, open-source, and commercial LLM backends for application orchestration.

  10. Overview of CSGHub LLM Asset Management

    CSGHub is an open-source asset management platform designed for the lifecycle of Large Language Models (LLMs). It allows users to govern datasets, model files, and code with support for private deployment and offline operation. It functions similarly to Hugging Face but provides private deployment options akin to GitLab (for source code) or Harbor (for container images).

  11. Overview of CSGHub Serverless API

    The CSGHub Serverless API allows developers to access models from the Transcendent Community for text, image, and speech processing tasks without managing deployment or infrastructure. It is designed for quick testing, prototyping, and experimentation, featuring GPU acceleration, automatic scaling, load balancing, and caching optimization.

  12. Overview of AI Gateway modules

    The AI Gateway is a unified AI interface management module in the CSGHub Admin Panel located in the left navigation bar below 'Computing Power Management'. It consists of four primary sub-modules:

    • Gateway Dashboard: Real-time monitoring of traffic, performance, and upstream health.
    • Public Inference: Management of self-hosted Serverless inference services.
    • Commercial API: Configuration and management of proxied third-party commercial model services.
    • MCP Gateway: Centralized registration and management of remote MCP Servers.