Skip to content

savantly-net/mainbot

Repository files navigation

mainbot

knowledge-explosion

Multi-tenant headless RAG chat bot.

This is a simple implementation of a multi-tenant chat bot that uses the Retrieval Augmented Generation.

Some of this code came from Wanderling, but all the game specific functions and rules-engine has been trimmed out (not ready to open source that one yet)

Quick start

Copy the example.env file to .env
Update the values for openai and pinecone.

make dev

open http://localhost:8080/swagger-ui/index.html

Overview

Features

  • Namespaced vector storage for multi-tenant preparation
  • OAuth authentication via JWT header
  • User session isolation
  • User session history and continuation
  • Adding documents to vector storage with a simple embedding pipeline
  • Retrieval Augmented Generation as default chat experience

Authentication Options

  • Anonymous
  • JWT
  • OAuth2

Conversation Storage Options

  • H2
  • Postgres

Vector storage Options

  • Pinecone
  • OpenSearch
  • In Memory

Embedding pipeline Options

  • OpenAI
  • OpenSearch
  • Replicate

Swagger

swagger

Starting a session

start a session

Sending a message

send a message

Example Configurations

Spring Profiles are used to configure the application.
See some example configurations (./src/main/resources/)[./src/main/resources/]

The configuration can be set in YAML, Properties, or Environment variables.

OpenAI for Embedding and Pinecone for Vector Storage

OPENAI_ENABLED=true
OPENAI_KEY=sk-xxxxxxxxxxxxxxxxxxxxxxxx
OPENAI_CHAT_MODEL_ID=gpt-3.5-turbo
PINECONE_ENABLED=true
PINECONE_API_KEY=xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
PINECONE_PROJECT_NAME=my-project
PINECONE_INDEX=my-index
PINECONE_ENVIRONMENT=us-east4-gcp
APP_SECURITY_ENABLED=false

OpenAI for Embedding and OpenSearch for Vector Storage

See the application configuration in application-opensearch.yml for more details.

SPRING_PROFILES_ACTIVE=opensearch
OPENAI_ENABLED=true
OPENAI_KEY=sk-xxxxxxxxxxxxxxxxxxxxxxxx
OPENAI_CHAT_MODEL_ID=gpt-3.5-turbo
OPENSEARCH_URL=https://localhost:9200

GitHub Action to upload

If you store documents in your repository, you can easily upload them to mainbot.
Leverage the GitHub Action to upload text/markdown files directly to mainbot.

name: Upload Files to Mainbot

on:
  workflow_dispatch: {} # Manually trigger the workflow
  push:
    branches:
      - main
    paths:
      - 'docs/**'
      - 'README.md'

jobs:
  upload:
    runs-on: ubuntu-latest
    steps:
      - name: Checkout repository
        uses: actions/checkout@v2
      
      - name: Upload Files
        uses: savantly-net/mainbot-github-action@main
        with:
          glob-patterns: '*.md'  # Set your file pattern here
          namespace: '/mainbot-documents'
          api-url: 'https://mainbot.my-company.apps.savantly.cloud'
          client-id: ${{ secrets.MAINBOT_CLIENT_ID }}
          client-secret: ${{ secrets.MAINBOT_CLIENT_SECRET }}
          token-endpoint: https://oidc.apps.savantly.cloud/realms/savantly/protocol/openid-connect/token

Backstage Integration

We've also built a Backstage plugin to integrate with mainbot.
Combining the Github action and the Techdocs plugin, you can make all of your documentation immediately available in Backstage and searchable in mainbot.

mianbot in backstage