search2ai

Simplified Chinese English

User communication

Telegram Channel

Friendly sponsorship

Version update

• V0.2.6, 20240425, supports SearXNG free search service, and supports Moonshot streaming mode in lossy
• V0.2.5, 20240425, In order to solve privacy concerns, the code of the open source search interface part
• V0.2.4, 20240424, supports Groq's llama-3, mistral and other models, take off at a speed
• V0.2.3, 20240423, Cloudflare Worker version supports Azure OpenAI; supports authorization code, and can customize user request keys.
• V0.2.2, 20240420, supports Moonshot's non-streaming mode
• V0.2.1, 20240310, supports Google, Bing, Duckduckgo, Search1API news searches; supports adjusting the number of search results through the environment variable MAX_RESULTS; supports adjusting the number of searches you want through the environment variable CRAWL_RESULTS
• V0.2.0, 20240310, optimized openai.js, cloudflare worker version, the speed is really faster this time!

For more historical updates, see Version Record

S2A

Let your big model API support networking, search, news, and web page summary. It has supported OpenAI, Gemini, and Moonshot (non-streaming). The big model will determine whether it is connected to the Internet based on your input. It does not search online every time. It does not require any plug-ins or replacement of keys. It can directly replace the custom address on your commonly used three-party client. It also supports self-deployment and will not affect other functions used, such as drawings, voice, etc.

Function

Model	Function	Streaming output	Deployment method
OpenAI	Internet connection, news, content crawling	Streaming, non-stream	Zeabur, on-premises deployment, Cloudflare Worker, Vercel
Azure OpenAI	Internet connection, news, content crawling	Streaming, non-stream	Cloudflare Worker
Groq	Internet connection, news, content crawling	Streaming, non-stream	Cloudflare Worker
Gemini	networking	Streaming, non-stream	Cloudflare Worker
Moonshot	Internet connection, news, content crawling	Partial flow, non-flow	Zeabur, local deployment, Cloudflare Worker (streaming), Vercel

use

Replace the client's custom domain name for your deployed address

deploy

Zeabur One-click deployment

Click the button to deploy in one click and modify environment variables

If you need to keep your project updated, it is recommended to fork this repository first and then deploy your branch through Zeabur

Local deployment

1. Cloning the repository to the local

 git clone https://github.com/fatwang2/search2ai

2. Copy .env.template to .env and configure environment variables
3. Enter the API directory, run the program, and display the log in real time

 cd api && nohup node index.js > output.log 2>&1 & tail -f output.log

4. Port 3014, the complete address after splicing is as follows. You can configure the APIBase address according to the client's needs (if https is needed, nginx is required for anti-generation, there are many online tutorials)

 http://localhost:3014/v1/chat/completions

Cloudflare Worker Deployment

1. Copy the code of search2openai.js, search2gemini.js or search2groq.js, without any modification! Deploy in cloudflare's worker. The address of the worker after it is launched can be used as the custom domain address when you call the interface. Pay attention to splicing. The worker address only represents the part before v1
2. Configuring environment variables in worker
3. Configure trigger-custom domain name in the worker. There may be problems with directly accessing the worker address in China, and it needs to be replaced with a custom domain name.

Vercel deployment

Special note: The vercel project does not support streaming output for the time being, and it has a 10s response limit. The actual user experience is not good. I mainly want to wait for the master to pull request for me.

One-click deployment

To ensure updates, you can also fork this project first and then deploy it on vercel yourself.

Environment variables

This project provides some additional configuration items, set through environment variables:

Environment variables	Is it necessary	describe	example
SEARCH_SERVICE	Yes	Your search service, what service you choose, what service key you need to configure	search1api, google, bing, serpapi, serper, duckduckgo, searxng
APIBASE	No	Third-party proxy address	https://api.openai.com, https://api.moonshot.cn, https://api.groq.com/openai
MAX_RESULTS	Yes	Number of search results	10
CRAWL_RESULTS	No	The number of in-depth searches (get the web page body after searching) is currently only supported for search1api, and the depth speed will be slow	1
SEARCH1API_KEY	No	If you choose search1api, I will build my own search service, which is fast and cheap. The application address is https://search1api.com	xxx
BING_KEY	No	If you choose Bing to search, please search for the tutorial yourself. The application address is https://search2ai.online/bing	xxx
GOOGLE_CX	No	If you select Google search, please search the tutorial yourself. The application address is https://search2ai.online/googlecx	xxx
GOOGLE_KEY	No	If you select Google search, API key, the application address is https://search2ai.online/googlekey	xxx
SERPAPI_KEY	No	If you choose serpapi, you can fill it in for 100 times/month. The registered address is https://search2ai.online/serpapi	xxx
SERPER_KEY	No	If you choose serper, the free quota for 2,500 times in 6 months is required. The registered address is https://search2ai.online/serper	xxx
SEARXNG_BASE_URL	No	If you choose the required serialxng, fill in the self-built serialXNG service domain name. Tutorial https://github.com/searxng/searxng, you need to open the json mode	https://search.xxx.xxx
OPENAI_TYPE	No	Openai supply source, default is openai	openai, azure
RESOURCE_NAME	No	Required if selecting Azure	xxxx
DEPLOY_NAME	No	Required if selecting Azure	gpt-35-turbo
API_VERSION	No	Required if selecting Azure	2024-02-15-preview
AZURE_API_KEY	No	Required if selecting Azure	xxxx
AUTH_KEYS	No	If you want the user to define the authorization code as the key separately when requesting, you need to fill it in. If you choose Azure, you must fill it in it.	000,1111,2222
OPENAI_API_KEY	No	If you want the user to define the authorization code as the key when requesting openai, you need to fill in it	sk-xxx

Subsequent iterations

• Fixed the Vercel project streaming output problem
• Improve streaming output speed
• Support more vertical searches