Amplifying GPT's capabilities by giving it access to locally executed plugins.
Why? So you can control what GPT should have access to: Access to parts of the local filesystem, allow it to access the internet, give it a docker container to use.
🚧 Under construction 🚧
The idea is for Auto-GPT, MemoryGPT, BabyAGI & co to be plugins for RunGPT, providing their capabilities and more together under one common framework.
Since the chat I/O is augmented, you as a user can also use the same features, for example include file contents in your prompt:
Let's continue implementing the chat functionality. Here is what we have so far:
{{ read_file("./lib/chat.ts") }}
Basically a locally run proxy to the OpenAI API with a plugin system and its own locally served chat UI.
Planned and mostly coded by GPT-4 itself. Development progress is tracked here:
👉 Development Tracking Issue #1
Check out development.md to see the current planning of the development as epics and stories and the prompts that created them.
- Install Deno if you haven't already.
- Clone this repository.
The RunGPT application requires the following permissions:
--allow-net: To make HTTP requests to the OpenAI API and listen for incoming connections to the local HTTP server.--allow-read: To read the API key from the environment and access local files.--allow-write: To create the plugins directory and write plugin files.--allow-env: To access the environment variable containing the API key.--allow-run: To run external commands, such asgit, for installing plugins.
Run the application with the required permissions:
deno run --allow-net --allow-read --allow-write --allow-env --allow-run rungpt.tsVisit http://localhost:8080 in your browser to use the web app.
make docker-build
make test