A simple AI "chat" shell (similar to ChatGPT) which can be used to "converse" with
one of the supported large language models, and optionally record your "chat transcript"
in a local file, making it easy to later reference your conversations with these models.
This app is designed to support access to multiple underlying chat models, including:
- OpenAI using Go OpenAI SDK
- Gemini using Google Gemini SDK
If you're not yet signed up to access either of these popular online language models, what are you waiting for? Clear instructions for obtaining your SDK Key needed to access each model on its "SDK" page using the link(s) above!
Copy the file config/config.yaml into config/config-local.yaml in order to create
your own, private configuration.
In your private configuration (i.e., config-local.yaml), select the model you wish to use,
and enter your credentials for accessing that model before running.
If you wish to create transcripts of your chat sessions, set the Scriber adapter value to
Template, and - if needed - customize its formatting in the TemplateScribe sections,
taking guidance from the go Templating language documentation.
You will need to have go installed in order to run gochat in your terminal. If you
don't already have this, see here to find out how.
In a go (version 1.21 or later) environment, run the app using the Makefile target, e.g.:
$ make run-chat
go run cmd/chat/main.go
Using model: gpt-4-1106-preview
Type 'exit' to quit
Ask me anything:
> Tell me in three words who you are.
Artificial Intelligence Assistant
> exit
Goodbye!
Bye bye!
$ Several envisioned enhancements for gochat in its Roadmap include:
- Add support for "chats" with supplied context, using an "embedding" service such as ChromaDB, Pinecone or Weaviate
- Add support for "chats" against confidential information, and/or without the need for internet connection, such as possible with ollama.
- Now that many LLMs (including Gemini) support media files (both as prompts and as responses), support for this paradigm could be really useful, and not so hard to implement.
- Envisioned is some sort of "meta-command" syntax to support access to local or remote media, TBD, ...
- Use something like Google's STT and TTS APIs to allow users to interact with the LLM using speech / audio.
- Please get in touch!
- Please submit a PR!