📖 llm-tracker

Search

SearchSearch

Code Assistants

Mar 24, 2024, 5 min read

We’ll probably move this somewhere else, but I figure it might be useful to put this in public somewhere since I’m researching Coding Assistants to help w/ a refactor of a largish code base.

I’m looking for practical tools for production use here, and less of the neat toys that’s popular atm.

See also:

Hosted Services

Cursor

GPT-4 backed VSCode-like editor

Sourcegraph Cody

Phind https://marketplace.visualstudio.com/items?itemName=phind.phind https://news.ycombinator.com/item?id=39471388 https://github.com/Lomusire/gpt4-unlimited-phin

https://continue.dev/ https://github.com/continuedev/continue

https://github.com/huggingface/llm-vscode

bloop

bloop is a developer assistant that uses GPT-4 to answer questions about your codebase. The agent searches both your local and remote repositories with natural language, regex and filtered queries.

  • YC Startup
  • Standalone GUI App (Tauri)
  • Github Access (OAuth)
  • Tantivy local search and Qdrnt local vector DB
  • Free version for individual use
    • Free version updated to only 10 uses per day! ($20/mo)
  • Uses OpenAI GPT-4, not possible to use your own API key

Mutable.ai

Supposedly will let you talk to your code base, create tests, review PRs, refactor folders, auto-document, etc…

OpenAI API

Code Interpreter

https://github.com/haseeb-heaven/code-interpreter

Instrukt

Coding AI: A coding assistant. Create indexes over any code base and attach it to the agent to do RAG (Retrieval Augmented Generation)

Maccarone

VSCode plugin that uses GPT-4 to write stubbed code snippets

Mentat

Mentat is an interactive command-line tool that can load and coordinate edits from your command line. It can directly edit files without copy and pasting (Y/n/i)

  • GPT-4 API
  • Shows tokens and costs for operations

GPT-4 Coding Assistant

GPT-4 Coding Assistant is a web application that leverages the power of OpenAI’s GPT-4 to help developers with their coding tasks. The application serves as an interactive chatbot that assists in code generation, understanding, and troubleshooting. It also utilizes embeddings and the Annoy library to search for similar code snippets in the provided codebase, offering more contextually relevant responses.

  • Uses OpenAI Embeddings

aider

aider is a command-line chat tool that allows you to write and edit code with OpenAI’s GPT models. You can ask GPT to help you start a new project, or modify code in your existing git repo. Aider makes it easy to git commit, diff & undo changes proposed by GPT without copy/pasting. It also has features that help GPT-4 understand and modify larger codebases.

  • Doesn’t have full embeddings or search, uses ctags

Chapyter

Chat w/ your Jupyter Lab code

CodeGPT

Write commit messages/does code review via git hook w/ OpenAI API

Local Code Interpreter

OpenAI ChatGPT-4 Code Interpreter

You get a VM sandbox and GPT-4 that knows how to run Python with it.

  • Limit of 50 exchanges/3 hours
  • VM will be recycled after a timeout period

Local Models

With the release of Meta’s Code Llama there is finally a model that is competitive with GPT-4 for code generation:

Also, it’s worth pointing out there was another model, a Llama2 70B full fine tune, Lemur that also performs quite well.

Note there are some API providers for those that can’t run these locally:

Local Apps

Refact

Refact is an open-source Copilot alternative available as a self-hosted or cloud option.

  • VS Code plugin
  • Cloud or self-hosted (docker container)
  • Uses AutoGPTQ/HF - supports WizardCoder for code, Llama2 for chat
  • Does not embed/search through code base yet

Continue

Continue is the open-source autopilot for software development—an IDE extension that brings the power of ChatGPT to VS Code and JetBrains

localpilot

Slick menubar interface to allow switching Github Copilot to use local models (built for Macs). Uses llama-cpp-python so should support any GGUF.

Code Llama for VSCode

An API which mocks llama.cpp to enable support for Code Llama with the Continue Visual Studio Code extension.

ggml p1

This project is an attempt to implement a local code completion engine utilizing large language models (LLM). Think of it as an open-source alternative to Github Copliot that runs on your device.

  • No code yet

(deprecated) TurboPilot

TurboPilot is a self-hosted copilot clone which uses GGML,the library behind llama.cpp, to run local code completion models. It was originally designed to run Salesforce codegen models but has recently been updated to provide support for Starcoder, Wizardcoder and most recently, StableCode Instruct from StabilityAI. It is heavily based and inspired by on the fauxpilot project.

  • Provides GPU inference via CUDA for Nvidia devices and OpenCL for Intel/AMD CPUS
  • Used to depend on a custom ggml fork but now tracks the ggerganov/ggml project.

localGPT

Index and search through local files w/ local models

  • Uses LangChain, Chroma, AutoGPTQ, llama.cpp

Other

Maybe useful, but not going to organize…

Code Interpreters

Toys

Backlinks

  • No backlinks found