Repository files navigation

Build your conversation-based search with LLM, support DeepResearch / DeepSeek R1.

Live Demo

Repositories:GitHubCNB

• 🔍New! Support "Deep Research" like OpenAI/Gemini/Perplexity.
• Built-in support for LLM: OpenAI, Google, Lepton, DeepSeek(R1), SiliconFlow, AliYun, Baidu, ChatGLM, Moonshot, Tencent, Lepton, Yi and more.
• Support Ollama,LMStudio
• Built-in support for search engine: Bing, Google,Tavily,SearXNG
• Customizable pretty UI interface
• Support light&dark mode/mobile
• Support i18n
• Support Continue Q&A with contexts.
• Support Cache results, Force reload.
• Support images search.

Support "Deep Research" like OpenAI/Gemini/Perplexity, through search engine, web scraping and LLM to iterate on any topic or question, and generate a comprehensive report. Project referencedeep-research, thanks to authordzhng.

Note:

• Warning: It will cost a lot of Tokens.
• Need to supportFunction Calling.
• Use JINA.ai to extract web page content (No need to configure KEY, limited: 20RPM).

Workflow:

1. Analyze user's query.
2. Generate follow-up questions to refine the research direction.
3. Generate and execute search queries.
4. Process and analyze search results.
5. Recursive exploration leads to deeper exploration base onstep 4.
6. Generate a comprehensive report.

More details can be found inDeepResearch discussion.

Install Docker.

docker pull docker.cnb.cool/aigc/aisearch

1.Get the code.

git clone https://github.com/yokingma/search_with_ai.gitcd search_with_ai

2.Edit.env.dockerfile. indeploy directory.

After modifying the .env.docker file, restart the Docker container to apply changes.

You must set at least one KEY.

...# OpenAI's keyOPENAI_KEY=#your key# Searxng hostname.SEARXNG_HOSTNAME=http://searxng:8080

3.Editmodel.jsonfile. [Optional]

{"provider":"openai","type":"openai","baseURL":"https://api.openai.com/v1","models": ["o1-preview","o1-mini","gpt-4o","gpt-4o-mini"]}

4.Run with Docker Compose.

docker compose up -d

Then visithttp://localhost:3000

5.Update

• Delete old images.
• Rundocker compose down
• Rundocker compose up -d

Built-in support for search engine: SearXNG, Bing, Google, Tavily, etc.

InstallSearXNG withsearxng-docker

Make sure to activate the json format to use the API. This can be done by adding the following line to the settings.yml file:

search:formats:        -html        -json

And set limiter to false:

server:limiter:false# default is true

apps/server/.env:

# SEARXNG_HOSTNAME=<host>

To use the Bing Web Search API, please visitthis link to obtain your Bing subscription key.

You have three options for Google Search: you can use the SearchApi Google Search API fromSearchApi,Serper Google Search API from Serper, or opt for theProgrammable Search Engine provided by Google.

Tavily is a search engine optimized for LLMs.

Jina Reader URL API, supporting full web content extraction. used in [DeepResearch] mode.JINA KEY is optional (limited to 20RPM).

# JINA API KEYJINA_KEY=#your key

• Node.js >= 20

• Turborepo

• PackageManager: yarn@3.5.1

• Directory Structure

apps/  | server # backend  | web # frontenddeploy/  | docker-compose.yaml # docker deployment file  | .env.docker # backend configuration file  | model.json # backend model configuration file  ...

• Development & BuildIn the root of the project:

turbo dev# orturbo build

• UpdateIn the root of the project:

git pull

This repository's source code is available under theMIT License.

🚀Back to top