The Power of ChatGPT

Every organisation has information that shouldn't leave its "walls". Beth's knowledge graph applies Subject Matter Constraints, allowing you to define which of its concepts can be passed to your language model and under what circumstances - and which can't.

Beth's knowledge graph also supports special concepts called dossiers that are only made available to the Beth with the permission of their owners. These can be used for customer, staff and other private data. They can be either posted to the graph using our API, or collected from your API when needed.

It may be years until you can plug in a Language Model that reliably reflects your organisation's values. Beth influences your model by mandating procedural knowledge graph entries for your policies, values and corporate "tone of voice". These are then used to instruct your plugged-in language model.

Large language models like ChatGPT are prone to "making things up" - particularly when the facts they need haven't been provided in their training data. Language models carry two broad classes of knowledge - procedures and facts. Beth trusts the model's procedural knowledge but constrains its use of facts to those that can be found in your knowledge graph.

Yes. If your user is has Editor permissions, they can use Workspace English to explicitly add facts to the knowledge graph. Beth may also use the history of comments made against each concept to augment the knowledge provided to any language model task that requests that concept.

If you've "plugged-in" your own language model, however, it may also learn from the information that Beth passes to it. You should always check this with the model's provider. Turnkey plan models don't get to keep your conversations and learn from them.

Turnkey plans use OpenAI's gpt-3.5-turbo (ChatGPT) model by default, though this may change as new models become available. The Enterprise plan also allows you use to plug in other models. Contact us to find out more.

Graph WorkSpace is a chat window on steroids. It's a place where you can add new concepts and facts to your knowledge graph using conversational English and simple forms. It also lets you test customer queries and fine-tune Beth's responses to them by modifying the your facts.

WorkSpace has visual tools that let you see the shape of your emerging knowledge graph and it maintains a conversational audit trail of who changed what.

Consumer channels are: Facebook Messenger, WhatsApp^*, Telegram, Discord and the beth.ai Chat Widget.

Enterprise channels are: Slack and Microsoft Teams.

* Coming soon

A query is any question or comment from a user in any connected chat device or in any Consumer or Enterprise channel. A request made using Graph Workspace or the REST API is also a query. Query counts are collected from all graphs that have been run during the month and then totalled.

Running your knowledge graphs "on-demand" is how we keep your costs down. It's a strategy that some call "serverless" computing. When a query arrives at your on demand graph, the graph is started to service that query. The graph then stays running while it's getting queries, and for a few more minutes after the last query is received.

Each graph runs on its own isolated graph database, so it can take a few seconds to start. However, it's usually less than the time it would take a human or a language model to type a response.

For Business and Enterprise plans, time-critical and high-use graphs can be set to "always on" in your Graphs Control Panel.

Yes you can. The Graphs Control Panel gives you comprehensive control over your account, its subscription plan and your graphs. Your plan may place some resource constraints on your graphs, but within those constraints you can control their resource footprint, their integrations and many other properties. For more detail see the FactNexus EKG Graphs Control Panel Reference.

Very. Your knowledge and data are encrypted at rest, and in transit. Enterprise users even get to control the key used for that encryption.

There are no multi-tenancy risks to your in-memory and working knowledge either. Each of your graphs runs on its own isolated graph database.

To register for an Essentials, Specialist, Workplace, Team, Business or Enterprise plan, first register for a Free Trial.

When you're ready, you can choose your plan by clicking Change Plan in the Account Details tab of your Graphs Control Panel.

Enterprise plan CPU-minutes, Workspace users and queries are paid for using beth.ai Credits. An Enterprise plan includes 500 credits per month. This is enough to cover the beth.ai resource needs of most enterprises. Credits that aren't used are accumulated for the next billing period and, if needed, more can be bought either manually or automatically.

It's a subscription strategy that gives you predictable costs and no risk of "bill shock".

Yes. You can build, modify and get answers from your graphs - and integrate them with other applications - using the  EKG REST API.

What's more, the API hides the structural complexity inherent in any knowledge graph by providing simple abstractions - Concepts, Facts, Cards and Dossiers.

From your Graphs Control Panel you can "bulk load" existing data into your knowledge graph. The product list from your WooCommerce, Squarespace, Shopify or other e-commerce store for example. Or the tags, titles and URLs of the articles or snippets from your Wordpress, Wix, Zendesk, Confluence or other Knowledge Base.

Prepare the table as a CSV (Comma Separated Value) file, then ingest it from your Graph Control Panel. Alternatively you can POST it using the REST API. For more detail see the FactNexus EKG Table-to-Graph API Reference.

Yes. Beth understands the semantics of a user's "I want to talk to a human" request and will take them to the relevant "handover" concept. It's then one click to connect with the live chat framework of your choice.

beth.ai is a powerful Knowledge Agent, but she's not designed for building procedural chatbot pathways. She does, however, have the semantic prowess to understand a user's intent.

To connect with your existing agent, you create a knowledge graph concept for each intent your procedural agent handles. Using a strategy we call "Intent Handover", Beth infers your user's intent then directs them to your bot via that concept.

See it working - ask Beth something like "Why is my messenger integration not working?"

Alternatively, you can use the REST API to collect responses from Beth and pipe them through your existing bot framework.

First up, ask Beth.

If Beth can't answer your question directly, you can ask to chat with a live agent.