English | 中文
GPT AI Assistant is a lightweight and extensible application that is implemented using the OpenAI API and LINE Messaging API.
Through the installation process, you can start to chat with your own AI assistant using the LINE mobile app.
- Log in to the OpenAI website.
- Generate an OpenAI API key.
- Log in to the LINE website.
- Add a provider (e.g. "My Provider").
- Create a channel (e.g. "My AI Assistant") of type Messaging API.
- Click the "Messaging API" tab and generate a channel access token.
- Log in to the GitHub website.
- Go to the
gpt-ai-assistant
project. - Click the "Star" button to support this project and the developer.
- Click the "Fork" button to copy the source code to your own repository.
- Go to the
- Log in to the Vercel website.
- Click the "Create a New Project" button to create a new project.
- Click the "Import" button to import the
gpt-ai-assistant
project. - Click the "Environment Variables" tab and add the following environment variables with their corresponding values:
OPENAI_API_KEY
with the OpenAI API key.LINE_CHANNEL_ACCESS_TOKEN
with the LINE channel access token.LINE_CHANNEL_SECRET
with the LINE channel secret.
- Click the "Deploy" button and wait for the deployment to complete.
- Click the "Domains" button to copy the application URL, e.g. "https://gpt-ai-assistant.vercel.app/".
- Go back to the LINE website.
- Go to the page of "My AI Assistant", click the "Messaging API" tab, set the "Webhook URL", e.g. "https://gpt-ai-assistant.vercel.app/webhook" and click the "Update" button.
- Click the "Verify" button to verify the webhook call is successful.
- Enable the "Use webhook" feature.
- Disable the "Auto-reply messages" feature.
- Disable the "Greeting messages" feature.
- Scan the QR code using the LINE mobile app to add as a friend.
- Start chatting with your own AI assistant!
On your own gpt-ai-assistant
project page, you can click on the "Sync fork" menu and then click on either the "Update branch" or "Discard commit" button to synchronize the latest code to your repository.
When the Vercel bot detects a change in the code, it will automatically redeploy.
Send commands using the LINE mobile app to perform specific functions.
Name | Description |
---|---|
version |
Check the application version. |
deploy |
Deploy the application. The VERCEL_DEPLOY_HOOK_URL environment variable is required. |
draw <prompt> |
Ask AI Assistant to draw an image. |
chat <prompt> |
Chat with AI Assistant. |
chat --auto-reply off |
Turn off auto-reply. The VERCEL_ACCESS_TOKEN environment variable is required. |
chat --auto-reply on |
Turn on auto-reply. The VERCEL_ACCESS_TOKEN environment variable is required. |
ai <prompt> |
An alias for the chat command. |
Set environment variables to change program settings.
Name | Default Value | Description |
---|---|---|
APP_DEBUG |
false |
Print prompt to console. |
APP_WEBHOOK_PATH |
/webhook |
Custom webhook URL path of application |
VERCEL_ACCESS_TOKEN |
null |
Vercel access token |
VERCEL_DEPLOY_HOOK_URL |
null |
Vercel deploy hook URL |
OPENAI_API_KEY |
null |
OpenAI API key |
OPENAI_COMPLETION_INIT_LANG |
zh |
Initial language. Must be one of zh , en or ja . |
OPENAI_COMPLETION_MODEL |
text-davinci-003 |
Refer to model parameter for details. |
OPENAI_COMPLETION_TEMPERATURE |
0.9 |
Refer to temperature parameter for details. |
OPENAI_COMPLETION_MAX_TOKENS |
240 |
Refer to max_tokens parameter for details. |
OPENAI_COMPLETION_FREQUENCY_PENALTY |
0 |
Refer to frequency_penalty parameter for details. |
OPENAI_COMPLETION_PRESENCE_PENALTY |
0.6 |
Refer to presence_penalty parameter for details. |
LINE_CHANNEL_ACCESS_TOKEN |
null |
LINE channel access token |
LINE_CHANNEL_SECRET |
null |
LINE channel secret |
Click the "Redeploy" button to redeploy if there are any changes.
- Check if the environment variables of the project are filled out correctly in the Vercel.
- Click the "Redeploy" button to redeploy if there are any changes.
- If there is still a problem, please go to Issues page, describe your problem and attach a screenshot.
Clone the project.
git clone [email protected]:memochou1993/gpt-ai-assistant.git
Go to the project directory.
cd gpt-ai-assistant
Install dependencies.
npm ci
Copy .env.example
to .env
.
cp .env.example .env
Set the environment variables as follows:
APP_ENV=local
APP_DEBUG=true
APP_PORT=3000
# optional
# VERCEL_GIT_REPO_SLUG=gpt-ai-assistant
# VERCEL_ACCESS_TOKEN=<your_vercel_access_token>
# VERCEL_DEPLOY_HOOK_URL=<your_vercel_deploy_hook_url>
# optional
# OPENAI_API_KEY=<your_openai_api_key>
# optional
# LINE_CHANNEL_ACCESS_TOKEN=<your_line_channel_access_token>
# LINE_CHANNEL_SECRET=<your_line_channel_secret>
Run the tests.
npm run test
Check the results.
> [email protected] test
> jest
console.info
=== 000000 ===
AI: 嗨!我可以怎麼幫助你?
Human: 嗨?
AI:
at Assistant.info [as debug] (assistant/assistant.js:55:28)
PASS assistant/index.test.js
✓ assistant works (1689 ms)
Test Suites: 1 passed, 1 total
Tests: 1 passed, 1 total
Snapshots: 0 total
Time: 2.579 s, estimated 4 s
Ran all test suites.
Set the environment variables as follows:
APP_ENV=production
Start a local server.
npm run dev
Start a proxy server.
ngrok http 3000
Go back to the website, modify the "Webhook URL" to e.g. "https://0000-0000-0000.jp.ngrok.io" and click the "Update" button.
Send a message from the LINE mobile app.
Check the results.
> [email protected] dev
> node api/index.js
=== 0x1234 ===
AI: 哈囉!
Human: 嗨?
AI: 很高興見到你!有什麼可以為你服務的嗎?
Start a local server.
npm run dev
Send a dummy request to the local server.
curl --request POST \
--url http://localhost:3000/webhook \
--header 'Content-Type: application/json' \
--data '{
"events": [
{
"type": "message",
"source": {
"type": "user",
"userId": "000000"
},
"message": {
"type": "text",
"text": "我是誰"
}
}
]
}'
Check the results.
> [email protected] dev
> node api/index.js
=== 000000 ===
AI: 嗨!我可以怎麼幫助你?
Human: 我是誰?
AI:
Detailed changes for each release are documented in the release notes.