Ghost is a small footprint AI assistant, built to run locally in your computer. It is meant for developers and can be modified/customised easily to suit your needs. Ghost is built using Langchain.

Ghost AI code assistant that runs on your laptop

Ghost can query any relational database, given the connection URL

Ghost can analyse CSV files on your laptop

Ghost can access different documents, search for images and read from images too

Ghost is written using Langchain and has configured the following tools:

1. Python REPL
2. Shell
3. Search (DuckDuckGo)
4. Ask Document (Unstructured File)
5. Ask CSV (Pandas)
6. Ask Database
7. Requests
8. Image Search (Google Image Search through SerpAPI) -- requires SerpAPI API key (https://serpapi.com)

Please check the tools.py file for more details.

It is fully customisable and extensible. In fact, while Ghost can be used out-of-the-box, it is meant to be customised for your own personal usage and style. Feel free to add or remove tools accordingly or change everything to what you want!

Setup

Dependencies

You need the following installed in your system:

• libpq-dev (ubuntu/debian) or libpq-devel (fedora/centos) or postgresql-libs (arch linux)
• tessdata (for OCR, make sure to set the path in the .env file)

After you have clone the repo, rename the .env.example file to .env and provide the necessary API key and other information from your provider.

Then set up the packages needed to run Ghost with pip.

$ pip install -r requirements.txt

You can consider setting up its own environment using Conda or pyenv or any of these similar tools.

Then you can run the app by doing this at the command line:

$ python ghost.py

Ghost specifications

You can change how Ghost behaves by changing specs.md or pointing the specs to a different file in the .env file. The specifications determine how Ghost will react to your prompts.

Google VertexAI

Ghost can also work with Google Cloud Platform (GCP) Vertex AI PaLM. GCP offers an enterprise version of PaLM through Vertex AI. The following instructions assumes you already have an account that has the necessary credentials to use Vertex AI.

To use Google Cloud Platform Vertex AI PaLM you can do the following:

1. If you don't already have the gcloud command-line interface (CLI), you can install it by following instructions here https://cloud.google.com/sdk/docs/install
2. Once you have gcloud, run this command in the command line

$ gcloud auth application-default login

3. This will open up a browser for you to log into the account with the credentials to access Vertex AI. Follow the instructions and click through to login and approve.
4. Once you you're done, an application_default_credentials.json file wil be created at the default location. Ghost will know how to locate this file (it's just using Google's default location to find it) to log into Vertex.

Saving output

By default, Ghost saves all your output to markdown format in a file named output.md. You can change it in the .env file.