Skip to main content

Foreword

About CoStrict​

Core Features​

  • Microservices Architecture: Containerized distributed service architecture
  • AI Gateway Integration: Supports multiple large language model integrations
  • Identity Authentication System: Integrates Casdoor for enterprise-grade identity management
  • Intelligent Code Analysis: Provides code review, completion, optimization, and more
  • Scalable Design: Supports horizontal scaling and custom plugins

System Architecture​

┌─────────────────┐    ┌─────────────────┐    ┌─────────────────┐
│   VSCode Plugin  │────│   API Gateway   │────│  Backend Services│
│   (CoStrict)    │    │  (Apache APISIX) │    │  (Microservices) │
└─────────────────┘    └─────────────────┘    └─────────────────┘
                               │                        │
                        ┌─────────────────┐    ┌─────────────────┐
                        │   AI Gateway    │    │  Database Cluster│
                        │   (Higress)     │    │   (PostgreSQL)  │
                        └─────────────────┘    └─────────────────┘

System Requirements​

Model Requirements​

All core features of CoStrict rely on large language models. You need to prepare the following model services and ensure the model API interfaces are functioning properly:

1. Chat model     (provide a complete http://chat_model_ip:chat_model_port/v1/chat/completions interface)
2. Embedding model (provide a complete http://embedding_model_ip:embedding_model_port/v1/embeddings interface)
3. Rerank model   (provide a complete http://rerank_model_ip:rerank_model_port/v1/rerank interface)
4. Completion model (provide a complete http://completion_model_ip:completion_model_port/v1/completions interface)

Note: Provide and record accurate model name, API KEY, and context length information for use during service deployment configuration.

Recommended Models and Download Links:

  • Chat model: GLM-4.6-FP8 or above, e.g. GLM-4.7, GLM-5

  • Embedding model: gte-modernbert-baseRAG/Embedding

  • Rerank model: gte-reranker-modernbert-baseRAG/Rerank

  • Download links:

https://modelscope.cn/models/ZhipuAI/GLM-4.7-FP8
https://modelscope.cn/models/iic/gte-modernbert-base
https://modelscope.cn/models/iic/gte-reranker-modernbert-base

Recommended Model Deployment Resources:

  • Chat model: 4 * H20 (recommended for GLM-4.7-FP8; double for GLM-4.5-FP8)

  • Embedding model: 0.5 * H20 or 0.5 * RTX4090

  • Rerank model: 0.5 * H20 or 0.5 * RTX4090

Trial Reminder:

  • If you have resources available, to experience the full feature set, please ensure all models meet the above requirements when deployed.
  • If you do not have resources, we offer two alternatives:
    • Use our officially released CoStrict directly without any additional deployment to experience all CoStrict features.
    • We can provide a time-limited online chat model API for short-term experience of CoStrict's primary AGENT and CODE REVIEW features.
FeatureSelf-deployed (models meeting requirements)Official CoStrict ReleaseUsing Time-limited API
AGENT (Vibe)✅ Full features✅ Full features✅ Time-limited experience (no Codebase)
AGENT (Strict)✅ Full features✅ Full features✅ Time-limited experience (no Codebase)
CODE REVIEW✅ Full features✅ Full features✅ Time-limited experience
Code Completion✅ Full features✅ Full features❌ Not supported
CoStrict Online API Access✅ No CoStrict online API access required❌ Requires CoStrict online API access❌ Requires CoStrict online API access

Self-deployed Backend Service Instance Environment​

Hardware Requirements:

  • CPU: Intel x64 architecture, minimum 16 cores (for Arm architecture deployment, please contact the CoStrict team)
  • Memory: Minimum 32GB RAM
  • Storage: Minimum 512GB available storage (note: this is runtime space required, including the Docker runtime environment and data generated during service operation)

Software Requirements:

  • OS: CentOS 7+ or Ubuntu 20.04+ (WSL supported)
  • Container Runtime: Docker 20.10+ (refer to offline Docker installation for offline setup)
  • Orchestration Tool: Docker Compose 2.0+