This is a tool I made to help HashiCorp Solutions Engineers in their day-to-day work. It supports the following use cases:
- Helping to write self-reflections, Feature Requests or generally improve writing.
- Perform RAG on custom documents. Please see inside
./docs
for further instructions. - Perform RAG on public website information, created via the
run_embed.sh
script.
Mac with M1 Pro
- Ollama - https://ollama.com/
- Docker
- Github access, to clone this repo
- Python 3.11
- Clone the repo
- Go to https://ollama.com/library and download some models. For example:
ollama pull mistral:7b
ollama pull llama3:8b
- Have Ollama running -> see
run_ollama.sh
. - Set
OLLAMA_HOST
indocker-compose.yaml
to the IP of your machine. - Run
docker compose up -d
.
Note: Once I have a docker container published, this will be easier
- Have Ollama running -> see
run_ollama.sh
- Run from source with
run_streamlit.sh
You can drop multiple files into the ./docs
folder and the app will load and create temporary in-memory embeddings for them, when loading the LLM. Currently supported file extensions are:
- csv
- doc or docx
- ppt or pptx
- xls or xlsx
- md or mdx
The quality of the answers only goes as far as the quality of the LLM, prompt and embeddings.
For RAG, create embeddings with the run_embed.sh
script or, if you work for HashiCorp, contact me directly.
See run_embed.sh
and edit the environment variables accordingly.