Summary

CoStrict is an open-source, enterprise-focused AI programming assistant designed for quality and control. It offers private deployment, strict AI development processes, and features like AI code review and completion, aiming to enhance enterprise software development workflows.

How It Works

CoStrict employs a "Strict Mode" that systematically breaks down requirements into stages like analysis, design, planning, and testing, ensuring standardized, high-quality AI output. Its Code Review functionality leverages company-wide RAG by indexing code repositories and uses a multi-expert model strategy with cross-confirmation for comprehensive checks on functions, files, or entire projects.

Quick Start & Requirements

CoStrict is available as a CLI tool and a JetBrains IDE plugin. The backend service is deployed using Docker Compose, requiring Docker 20.10+ and Docker Compose 2.0+. Specific hardware includes a minimum of 16 CPU cores, 32GB RAM, and 512GB storage for the backend service. Crucially, it requires access to four types of large language model services: chat, embedding, rerank, and completion, each with specific API interface requirements and recommended models (e.g., GLM-4.6-FP8, Qwen3-4B-Instruct-2507). Recommended model deployment resources range from 0.5 to 4x H20 or RTX4090 GPUs per model type.

• Deployment Guide: https://docs.costrict.ai/en/deployment/introduction

Highlighted Details

• Strict Mode: Standardizes AI code generation through a structured, multi-step process.
• Code Review: Implements company-wide RAG for coding knowledge with multi-expert model validation.
• Vibe Mode: Facilitates rapid development and simple tasks via natural language dialogue.
• MCP Service: Enables integration with external APIs, databases, and custom tools.
• Model Flexibility: Supports multiple free advanced models, third-party APIs (Anthropic, OpenAI), and local models (Ollama).
• Contextual Awareness: Dynamically incorporates repository data, cursor position, and allows manual context additions (files, terminals, issues) via the '@' key.
• OpenSpec Integration: Facilitates AI agent workflows for change proposals, feature planning, implementation, and review.
• Image Context: Supports image uploads and drag-and-drop for context.

Maintenance & Community

Contributions are welcomed via GitHub Issues and pull requests, following the "How To Contribute" guidelines. Community communication primarily occurs through GitHub. Key acknowledgments include Roo Code, OpenSpec, agents.md, and agentskills. The project is actively maintained by Sangfor, Inc.

Licensing & Compatibility

CoStrict is licensed under the Apache 2.0 license. Its design emphasizes private deployment for enterprises, ensuring code security through physical isolation and end-to-end encryption, making it compatible with closed-source environments requiring data privacy.

Limitations & Caveats

The project includes a comprehensive disclaimer from Sangfor, Inc., stating that all tools and outputs are provided "AS IS" and "AS AVAILABLE." Users assume all risks, including potential intellectual property infringement, cyber vulnerabilities, bias, inaccuracies, errors, defects, viruses, and property/personal injury. Users are solely responsible for the legality, appropriateness, and results of their use. The setup requires significant LLM infrastructure and configuration.