@7-docs/cli

7d is a powerful CLI tool to ingest content and store into a vector database, ready to get queried like you would with ChatGPT.

Uses OpenAI APIs, part of 7-docs.

Impression

CLI

Content

• Status
• Prerequisites
• Installation
• Pinecone
• Supabase
• Ingestion
  • Local
  • GitHub
  • HTTP
  • PDF

Status

This is still in the early days, but already offers a variety of features:

• Plain text, Markdown and PDF files are supported as input.
• Ingest from local files, from HTML pages over HTTP, and from GitHub repositories.
• The OpenAI text-embedding-ada-002 model is used to create embeddings.
• Pinecone and Supabase are supported for vector storage.
• The OpenAI gpt-3.5-turbo model is used for chat completions from the CLI.

See the 7-docs overview for more packages and starter kits.

Prerequisites

• Node.js v16+
• OpenAI API key
• Pinecone or Supabase account, plus API keys
• When ingesting lots of files from GitHub, a GitHub token

Installation

You can install 7-docs in two ways:

• Global to manage knowledge base(s) from the command line.
• Local to manage the knowledge base(s) of a repository.

Global

Use 7d from anywhere to manage your personal knowledge bases:

npm install --global 7-docs

Get an OpenAI API key and make it available as en environment variable:

export OPENAI_API_KEY=sk-xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

Alternative storage (in ~/.7d.json) so it's available in your next session too:

7d set OPENAI_API_KEY sk-xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

This works for the other export values shown later as well.

Local

Add 7d to the devDependencies of a repository to manage its knowledge base(s):

npm install --save-dev 7-docs

Store the variables you need in a local .env file in the root of your project:

OPENAI_API_KEY="sk-xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx"

For local installations, use npx 7d (over just 7d).

Now let's choose either Pinecone or Supabase!

Pinecone

Make sure to have a Pinecone account and set PINECONE_URL and PINECONE_API_KEY:

export PINECONE_URL=xxxxx-xxxxxxx.svc.us-xxxxx-gcp.pinecone.io
export PINECONE_API_KEY=xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx

Create or select an index:

7d pinecone-create-index --index my-index

Supabase

Make sure to have a Supabase account and set SUPABASE_URL and SUPABASE_API_KEY:

export SUPABASE_URL="https://xxxxxxxxxxxxxxxxxxxx.supabase.co"
export SUPABASE_API_KEY="ey..."

Print the SQL query to enable pgvector and create a table (paste the output in the Supabase web admin):

7d supabase-create-table --namespace my-collection

Ingestion

Let's ingest some text or Markdown files, make sure to adjust the --files pattern to match yours:

7d ingest --files README.md --files 'docs/**/*.md' --namespace my-collection

Note that ingestion from remote resources (GitHub and/or HTTP) has the benefit to link back to the original source when retrieving answers. This is not possible when using local files.

GitHub

Use --source github and file patterns to ingest from a GitHub repo:

7d ingest --source github --repo reactjs/react.dev --files 'src/content/reference/react/*.md' --namespace react

You can start without it, but once you start fetching lots of files you'll need to set GITHUB_TOKEN:

export GITHUB_TOKEN="ghp_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx"

HTTP

Crawl content from web pages:

7d ingest --source http --url https://en.wikipedia.org/wiki/Butterfly

PDF

7d supports PDF files as well:

7d ingest --files ./my-article.pdf
7d ingest --source github --repo webpro/webpro.nl --files 'content/*.pdf'

Query

Now you can start asking questions about it:

7d Can you please give me a summary?

Other commands

Other convenience flags and commands not mentioned yet.

--help

Shows available commands and how they can be used:

7d --help

openai-list-models

List available OpenAI models:

7d openai-list-models

pinecone-clear-namespace

Clear a single namespace from the current Pinecone index:

7d pinecone-clear-namespace --namespace my-collection

Token Usage

The OpenAI recommendation text-embedding-ada-002 model is used to create embeddings. Ingestion uses some tokens when ingesting lots of files. Queries use only a few tokens (using the gpt-3.5-turbo model by default). See the console for details.