Backend Deployment

Project Overview

CoStrict Backend Deployment Tool is an enterprise-level AI code assistant backend service deployment solution based on Docker Compose. This project provides a complete microservice architecture, including core components such as AI gateway, identity authentication, code analysis, and chat services, supporting both private deployment and cloud service modes.

To view the deployment architecture, server requirements, environment requirements, model requirements, model download links, etc, please visit Foreword

Quick Start

1. Get Deployment Configuration

Method 1: Via GitHub Release

Visit the following address:

https://github.com/zgsm-sangfor/costrict-deploy-docker/releases/

Download the latest release archive costrict-backend-deploy-vX.X.X.tar.gz, e.g. costrict-backend-deploy-v0.0.2.tar.gz

Copy the archive to the server, then extract it:

# costrict-server is the deployment directory, please remember this directory as your deployment directory, it will be used later
# This directory will store most runtime data, please ensure stable and sufficient disk space
mkdir ./costrict-server
# Extract all files into costrict-server; replace costrict-backend-deploy-v0.0.2.tar.gz with the actual version you downloaded
tar -zxf costrict-backend-deploy-v0.0.2.tar.gz -C costrict-server
# Enter the deployment directory
cd ./costrict-server

Note, ./costrict-server is the deployment directory,

2. Prepare Backend Service Images

Note: if you are deploying in an offline environment, continue reading this section. If your server can access the full internet (including docker.io, quay.io, docker.elastic.co, etc.), you can skip this step — all images will be pulled automatically during deployment.

The images required by the CoStrict backend can be found in the scripts/newest-images.list file for a complete list. You can also pull these images manually.

We provide a Baidu Netdisk download link:

Netdisk address:

https://pan.baidu.com/s/5H0ppvaTja4g2MKZs0Ki1-g

Newest Version is: v0.0.3

After downloading all tar packages and copying them to a directory on the server (e.g. /root/images), run:

# /root/images is the directory containing the tar packages; please replace accordingly
# scripts/load-images.sh is the script in the deployment directory, please find it yourself
bash scripts/load-images.sh -l /root/images

### 3. Environment Configuration

Edit the configuration file (Note: edit, not create new, it's in the deployment directory and already contains some content):

```bash
vim configure.sh

Key Configuration Parameters:

For a quick start, you only need to configure one parameter — the server's IP address. The CoStrict client will use this IP to access the CoStrict backend services. Be sure to set this parameter before continuing:

# Note, this configuration is on the first line, just edit it directly
COSTRICT_BACKEND=""

4. Service Deployment

A single command is all it takes to bring up all CoStrict services:

bash costrict.sh install

When the process completes, output similar to the following will be displayed — save it somewhere:

[INFO]  Admin user access (casdoor): http://192.168.79.130:39009/
[INFO]  Configure Chat model at (nacos): http://192.168.79.130:31808/
[INFO]  Set BaseUrl to: http://192.168.79.130:39080/

Service Configuration

Model Configuration

Higress has been removed in the current version. To configure models, please go to nacos, e.g. http://192.168.79.130:31808/ (use the actual URL from your output).

Username: nacos
Password: nacos

Open Configuration Management → Configuration List → costrict (below the "Configuration Management" heading).
Find the model-config entry, click Edit, and modify the Configuration Content according to the instructions — do not touch anything else.
Publish after editing; changes take effect in real time automatically.

The following is the configuration reference/example. Note:

Please follow YAML or JSON format strictly.

When publishing, nacos performs a format check. If the error Configuration may have syntax errors, are you sure you want to submit? appears, do not submit.

# Configuration example and explanation — edit this in nacos, do not copy this config directly.
models:
    # Example 1
    # id: the model name visible in the plugin / model list — freely customizable
  - id: Kimi-K2-Moonshot
    # Model information
    publicInfo:
      # Max output tokens, must be smaller than contextWindow (32K is usually sufficient). Use Arabic numerals.
      maxTokens: 32000
      # Context window size provided by your model provider. If the value is less than 16000, switch to a larger model. Use Arabic numerals.
      contextWindow: 200000
    # Model key settings
    privateConfig:
      # The real model name provided by your model provider
      convertedName: KimiK1000000
      # Full endpoint path /v1/chat/completions
      addr: http://127.0.0.1:6666/v1/chat/completions
      # Custom headers — add as many as needed
      extraHeaders:
        # Auth header usually goes here; ask your model provider for the value
        Authorization: "sk-xxxxxxxx"
      # Whether to skip SSL verification. Only needed when the service uses HTTPS with a self-signed certificate. Ask your model provider.
      skipSSLVerify: false
    # Example 2 — same as Example 1 but with more supplementary notes.
    # Model 2: add as many models as you need, following YAML format strictly.
  - id: Qwen3.5
    # Detailed model configuration options
    publicInfo:
      object: model
      maxTokens: 32000
      contextWindow: 200000
      # Whether image input is supported
      supportsImages: true
      supportsComputerUse: false
      supportsPromptCache: false
      supportsReasoningBudget: false
      requiredReasoningBudget: false
      # Description
      description: Qwen3.5 to glm
      # Quota — this feature is not currently active.
      creditConsumption: 3
      creditDiscount: 1
    privateConfig:
      convertedName: glm-5-tokens
      addr: http://127.0.0.1:32323/v1/chat/completions
      extraHeaders:
        Authorization: sk-***
      skipSSLVerify: false

Optional: Identity Authentication System Configuration (Casdoor)

If there are no special requirements (e.g., integrating with third-party authentication systems), please skip this step, skip this step, skip this step.

For the first testing trial, also skip this step, skip this step, skip this step.

Access the Casdoor management interface at the following address:

# The first URL shown after installation completes
http://{COSTRICT_BACKEND}:{PORT_CASDOOR}

Basic user setup reference: Casdoor Basic Settings

Client Integration

CoStrict Plugin Configuration

Install the CoStrict VSCode extension
Open the "Provider" page in the extension settings
Select the API provider as "CoStrict"

Configure the backend service address:

CoStrict Base URL: {COSTRICT_BACKEND_BASEURL}

Click "Login CoStrict" to complete authentication

Test login credentials:

Username: costrict
Password: 123

Detailed installation guide: CoStrict Download and Installation Documentation (includes VSCode and JetBrains IDE)

Service access address:

# The third URL shown after installation completes
Default backend entry: http://{COSTRICT_BACKEND}:{PORT_APISIX_ENTRY}

Others

More information

CoStrict - Let AI power your code development journey

Project Overview​

Quick Start​

1. Get Deployment Configuration​

2. Prepare Backend Service Images​

4. Service Deployment​

Service Configuration​

Model Configuration​

Optional: Identity Authentication System Configuration (Casdoor)​

Client Integration​

CoStrict Plugin Configuration​

Others​