Scriv

About

Scriv is a platform for creating custom chatbots with deep knowledge of any data. It uses retrieval augmented generation (RAG) to quickly find relevant information from websites, uploaded documents, company wikis, and connected Slack communities and use it to provide domain-specific, context-aware chatbots. These bots can be used by internal corporate teams or installed on websites for public-facing knowledge bots.

As a codebase, Scriv is the perfect launching point for building a B2B application on top of LLM technologies. It shows how you can build a production, multi-tenant LLM application on top of SaaS Pegasus, showcasing many important features and tools.

Highlights

• A complete RAG pipeline, built on top of LangChain, pgvector, OpenAI embeddings, and GPT 3.5/4/4o. It supports prompt customization and history.
• An extremely flexible data model that supports documents, collections, bots, and teams—allowing companies to easily create different bots for different parts of their knowledge bases.
• A two-way Slack integration that lets customers talk to bots in Slack, and indexes Slack information as part of the knowledge base. This integration could be extended to other platforms.
• An embeddable chat widget component that can be used to install the bot on any website, with custom styling.
• Asynchronous, background operations for working with files, indexing knowledge bases, and responding to chat messages.

Technologies

• Back end: Python 3.11, Django 4.2 LTS, Postgres, Celery (tasks)
• Front end: TailwindCSS, HTMX
• AI: LangChain, OpenAI, pgvector
• Deployment: fly.io (can be changed)

What’s included

• Complete source code for https://scriv.ai/, https://docs.scriv.ai/ and the chat widget.
• Email/Slack support for setting up the code.

What’s not included

• The scriv.ai domain and marketing assets.
• The content of the scriv.ai guides.