https://github.com/different-ai/embedbase-internet-search
An internet search extension for Embedbase
https://github.com/different-ai/embedbase-internet-search
Last synced: 4 months ago
JSON representation
An internet search extension for Embedbase
- Host: GitHub
- URL: https://github.com/different-ai/embedbase-internet-search
- Owner: different-ai
- License: mit
- Created: 2023-05-22T08:39:58.000Z (almost 2 years ago)
- Default Branch: main
- Last Pushed: 2023-05-26T14:06:17.000Z (almost 2 years ago)
- Last Synced: 2024-11-13T23:48:21.069Z (5 months ago)
- Language: Python
- Homepage:
- Size: 206 KB
- Stars: 3
- Watchers: 3
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
- jimsghstars - different-ai/embedbase-internet-search - An internet search extension for Embedbase (Python)
README
# embedbase-internet-search
embedbase-internet-search - internet search extension for [Embedbase](https://github.com/different-ai/embedbase)
⚠️ Status: Alpha release ⚠️
[](https://github.com/psf/black)
[](https://github.com/different-ai/embedbase-internet-search/blob/main/LICENSE)
The point of internet search in embedbase is to combine your private information with latest public information.
Also remember that AIs like ChatGPT have limited knowledge to a certain date, for example try to ask ChatGPT about GPT4 or about Sam Altman talk with the senate (which happened few days ago), it will not know about it.
https://github.com/different-ai/embedbase-internet-search/assets/25003283/3131450f-93e2-46b1-8dcb-30673e5abe70
Please check [examples](./examples/answer-question/README.md) for usage or keep reading.
## Quick tour
Here's an example to answer general purpose questions using embedbase hosted:
The recommended workflow is like this:
1. search your question using internet endpoint
2. (optional) add results to embedbase
3. (optional) search embedbase with the question
4. use `.generate()` to get your question answered
```ts
// npm i embedbase-js
import { createClient } from 'embedbase-js'
const formatInternetResultsInPrompt = (internetResult: any) =>
`Name: ${internetResult.title}
Snippet: ${internetResult.snippet}
Url: ${internetResult.url}`
const system = `You are an AI assistant that can answer questions.
When a user send a question, we will answer its question following these steps:
1. we will search the internet with the user's question.
2. we will ask you to answer the question based on the internet results.`
const fn = async () => {
const embedbase = createClient('https://api.embedbase.xyz', process.env.EMBEDBASE_API_KEY)
// get question from process.argv
const question = process.argv[2]
if (!question) {
console.log('Please provide a question as argument, for example "What is GPT4?"')
return
}
const results = await embedbase.internetSearch(question)
const answerQuestionPrompt = `Based on the following internet search results:
${results.map(formatInternetResultsInPrompt).join('\n')}
\n
Please answer the question: ${question}`
for await (const result of embedbase.generate(answerQuestionPrompt, {
history: [{
role: 'system',
content: system
}],
})) {
console.log(result)
}
}
fn()
```
```bash
EMBEDBASE_API_KEY="" npx tsx answer.ts "What did Sam Altman say to to US Senate lately?"
```
## Self-hosted usage
Just add two lines to your original embedbase entrypoint:
```py
import os
import uvicorn
from embedbase import get_app
from embedbase.database.memory_db import MemoryDatabase
from embedbase.embedding.openai import OpenAI
# import this
from embedbase_internet_search import internet_search
app = get_app().use_db(MemoryDatabase()).use_embedder(OpenAI(os.environ["OPENAI_API_KEY"])).run()
# add the new endpoint
app.add_api_route("/internet-search", internet_search, methods=["POST"])
if __name__ == "__main__":
uvicorn.run(app)
```
[Check how it is used in Embedbase hosted](https://github.com/different-ai/embedbase/blob/4e1181eb8e162e1a9544bf72461a721d10f458fc/hosted/main.py#L38)
If you have any feedback or issues, please let us know by opening an issue or contacting us on [discord](https://discord.gg/pMNeuGrDky).
Regarding the SDK, please refer to the [documentation](https://docs.embedbase.xyz/sdk).