search

stitionai / devika

Devika is an Agentic AI Software Engineer that can understand high-level human instructions, break them down into steps, research relevant information, and write code to achieve the given objective. Devika aims to be a competitive open-source alternative to Devin by Cognition AI.
MIT License
17.88k stars 2.33k forks source link

Terminal/ASCII based browser #440

Open PlanetMacro opened 2 months ago

PlanetMacro commented 2 months ago

Is your feature request related to a problem?

Using Search Engine APIs can be expensive and you need to create accounts and hence share data and become third party dependent.

Describe the solution you'd like

ASCII/Terminal based browser can be used for normal web browsing / search engines without APIs or screenshots. LLMs are usually good in extracting relevant data from ASCII based browsers. Implement a lightweight ASCII based browser like W3M, Links2 or eLinks and such and make the machine use that for web search.

cafeTechne commented 2 months ago

This is a great idea! Can anyone give me an outline of how to do this?

On Sun, Apr 14, 2024 at 7:33 AM Robert Paulson @.***> wrote:

Is your feature request related to a problem?

Using Search Engine APIs can be expensive and you need to create accounts and hence share data and become third party dependent. Describe the solution you'd like

ASCII/Terminal based browser can be used for normal web browsing / search engines without APIs. LLMs are usually good in extracting relevant data from ASCII based browsers. Implement a lightweight ASCII based browser like W3M, Links2 or eLinks and such and make the machine use that for web search.

— Reply to this email directly, view it on GitHub https://github.com/stitionai/devika/issues/440, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFACJ5XYT4GB7XJZCW5KZH3Y5JSR7AVCNFSM6AAAAABGGEIZZ2VHI2DSMVQWIX3LMV43ASLTON2WKOZSGI2DEMJRGA2TGMA . You are receiving this because you are subscribed to this thread.Message ID: @.***>

PlanetMacro commented 2 months ago

If you are willing to spend 20-30 $ you can utilize the large token window of Claude 3 and use https://github.com/mufeedvh/code2prompt to upload the entire code base into claude 3 opus. From experience, it will not be able to tell you exactly what to do, but it will very likely put your awareness to the right parts of the code without you needed to learn the full code base.

The biggest obstacle would be to find a way that the LLM agent can use the ascii browser to open links from the ascii browser. This should ultimately be necessary though, to actively crawl the web.