ChatGPT Retrieval Plugin: Difference between revisions

Latest revision as of 23:29, 8 April 2023

See also: ChatGPT and ChatGPT Plugins

Introduction

The ChatGPT Retrieval Plugin is a powerful addition to OpenAI's ChatGPT, enabling the AI to access and utilize data stored in a vector database. This capability not only enhances ChatGPT's performance by providing access to customized data but also addresses its long-standing limitation of lacking long-term memory. The plugin is part of the broader ChatGPT plugins ecosystem, which allows ChatGPT to interact with various apps and services, transforming it from a conversation tool to an AI capable of taking actions in the real world.

Background Information

Recently, OpenAI introduced plugins for ChatGPT, which have significantly expanded the chatbot's capabilities. These plugins act as a bridge between the chatbot and a range of third-party resources, enabling it to leverage these resources to perform tasks based on user conversations. The Retrieval Plugin, specifically, has the potential to be widely used, as it allows users to create customized versions of ChatGPT tailored to their own data.

With the integration of plugins, ChatGPT can now perform various tasks such as ordering groceries, booking restaurants, and organizing vacations by utilizing services like Instacart, OpenTable, and Expedia. Moreover, the Zapier plugin allows ChatGPT to connect with thousands of other applications, from Google Sheets to Salesforce, thus broadening its reach.

ChatGPT Retrieval Plugin: Connecting to a Vector Database

The ChatGPT Retrieval Plugin enables users to connect ChatGPT to an instance of a vector database, allowing any information stored in the connected database to be used to answer questions and provide responses based on the details stored in the database. Additionally, the vector database can be utilized as a long-term storage solution for ChatGPT, allowing it to persist and store portions of user conversations beyond the short-lived memory of a browser tab.

Plugin Functionality

The main functions of the ChatGPT Retrieval Plugin include:

Connecting a vector database with proprietary data to ChatGPT, allowing it to answer specific questions based on that data
Persisting personal documents and details to provide a personalized touch to ChatGPT's responses
Storing conversations with ChatGPT in the attached vector database, enabling continued conversations across multiple sessions

This functionality allows for regular updates to content stored in connected vector databases, giving the model awareness of new information without the need for costly and time-consuming retraining of the large language model (LLM).

Overview of Vector Database Providers

This article presents an overview of various vector database providers supported by the plugin, highlighting their unique features, performance, and pricing. The choice of a vector database provider depends on specific use cases and requirements, and each provider necessitates the use of distinct Dockerfiles and environment variables. Detailed instructions for setting up and using each provider can be found in their respective documentation at /docs/providers/<datastore_name>/setup.md.

Pinecone

Pinecone is a managed vector database engineered for rapid deployment, speed, and scalability. It uniquely supports hybrid search and is the sole datastore that natively accommodates SPLADE sparse vectors. For comprehensive setup guidance, refer to /docs/providers/pinecone/setup.md.

Weaviate

Weaviate is an open-source vector search engine designed to scale effortlessly to billions of data objects. Its out-of-the-box support for hybrid search makes it ideal for users who need efficient keyword searches. Weaviate can be self-hosted or managed, offering flexible deployment options. For extensive setup guidance, refer to /docs/providers/weaviate/setup.md.

Zilliz

Zilliz is a managed, cloud-native vector database tailored for billion-scale data. It boasts a plethora of features, including numerous indexing algorithms, distance metrics, scalar filtering, time-travel searches, rollback with snapshots, full RBAC, 99.9% uptime, separated storage and compute, and multi-language SDKs. For comprehensive setup guidance, refer to /docs/providers/zilliz/setup.md.

Milvus

Milvus is an open-source, cloud-native vector database that scales to billions of vectors. As the open-source variant of Zilliz, Milvus shares many features with it, such as various indexing algorithms, distance metrics, scalar filtering, time-travel searches, rollback with snapshots, multi-language SDKs, storage and compute separation, and cloud scalability. For extensive setup guidance, refer to /docs/providers/milvus/setup.md.

Qdrant

Qdrant is a vector database that can store documents and vector embeddings. It provides self-hosted and managed Qdrant Cloud deployment options, catering to users with diverse requirements. For comprehensive setup guidance, refer to /docs/providers/qdrant/setup.md.

Redis

Redis is a real-time data platform suitable for an array of applications, including AI/ML workloads and everyday use. By creating a Redis database with the Redis Stack docker container, Redis can be employed as a low-latency vector engine. For a hosted or managed solution, Redis Cloud is available. For extensive setup guidance, refer to /docs/providers/redis/setup.md.

LlamaIndex

LlamaIndex serves as a central interface to connect your LLMs with external data. It offers a collection of in-memory indices over structured and unstructured data for use with ChatGPT. Unlike conventional vector databases, LlamaIndex supports a wide array of indexing strategies (e.g., tree, keyword table, knowledge graph) optimized for various use-cases. It is lightweight, user-friendly, and requires no additional deployment. Users need only specify a few environment variables and, optionally, point to an existing saved Index JSON file. However, metadata filters in queries are not yet supported. For comprehensive setup guidance, refer to /docs/providers/llama/setup.md.

Weaviate Retrieval Plugin in Action

The Weaviate Retrieval Plugin can be used in various applications, such as creating a private, customized version of ChatGPT tailored to an organization's internal documents or personalizing ChatGPT based on individual user details. By connecting the plugin to Weaviate, users can make ChatGPT more useful and relevant to their specific needs.

Using ChatGPT on Proprietary Company Documents

The Weaviate Retrieval Plugin can be used to create a customized version of ChatGPT trained on a company's internal documents, enabling it to act as a human resources chatbot. This can provide employees with easy access to information about onboarding processes, health benefits, and more.

Personalizing ChatGPT

The plugin also enables the customization of ChatGPT around personal details, such as information about friends or languages spoken by the user. By storing these details in Weaviate, ChatGPT can provide more tailored and personalized responses.

Helping ChatGPT Remember

One of the most powerful applications of the Weaviate Retrieval Plugin is its ability to store and reference previous conversations with ChatGPT. By persisting these conversations in Weaviate, ChatGPT can recall past interactions and provide more contextually relevant responses.

@@ Line 1: / Line 1: @@
+{{see also|ChatGPT|ChatGPT Plugins}}
+[https://github.com/openai/chatgpt-retrieval-plugin GitHub]
 ==Introduction==
-The ChatGPT Retrieval Plugin is a powerful addition to OpenAI's ChatGPT, enabling the AI to access and utilize data stored in a vector database like Weaviate. This capability not only enhances ChatGPT's performance by providing access to customized data but also addresses its long-standing limitation of lacking long-term memory. The plugin is part of the broader ChatGPT plugins ecosystem, which allows ChatGPT to interact with various apps and services, transforming it from a conversation tool to an AI capable of taking actions in the real world.
+The [[ChatGPT Retrieval Plugin]] is a powerful addition to [[OpenAI]]'s [[ChatGPT]], enabling the [[AI]] to access and utilize data stored in a [[vector database]]. This capability not only enhances ChatGPT's performance by providing access to customized data but also addresses its long-standing limitation of lacking long-term memory. The plugin is part of the broader [[ChatGPT plugins]] ecosystem, which allows ChatGPT to interact with various apps and services, transforming it from a conversation tool to an AI capable of taking actions in the real world.
 ==Background Information==
-Recently, OpenAI introduced plugins for ChatGPT, which have significantly expanded the chatbot's capabilities. These plugins are more than just the "eyes and ears" of ChatGPT as they provide the AI with access to third-party resources and enable it to act on the information obtained from conversations.
+Recently, OpenAI introduced plugins for ChatGPT, which have significantly expanded the [[chatbot]]'s capabilities. These plugins act as a bridge between the chatbot and a range of third-party resources, enabling it to leverage these resources to perform tasks based on user conversations. The Retrieval Plugin, specifically, has the potential to be widely used, as it allows users to create customized versions of ChatGPT tailored to their own data.
-===ChatGPT Plugins in Action===
+With the integration of plugins, ChatGPT can now perform various tasks such as ordering groceries, booking restaurants, and organizing vacations by utilizing services like [[Instacart]], [[OpenTable]], and [[Expedia]]. Moreover, the Zapier plugin allows ChatGPT to connect with thousands of other applications, from [[Google Sheets]] to [[Salesforce]], thus broadening its reach.
-With the integration of plugins, ChatGPT can now perform various tasks such as ordering groceries, booking restaurants, and organizing vacations by utilizing services like Instacart, OpenTable, and Expedia. Moreover, the Zapier plugin allows ChatGPT to connect with thousands of other applications, from Google Sheets to Salesforce, thus broadening its reach.
-Among these plugins, the Retrieval Plugin stands out due to its ability to overcome ChatGPT's lack of long-term memory and provide responses based on internal documents and data. This feature is particularly valuable for users who wish to create customized versions of ChatGPT tailored to their specific data.
+==ChatGPT Retrieval Plugin: Connecting to a Vector Database==
+[[File:chatgpt retrieval plugin1.png|400px|right]]
+The [[ChatGPT Retrieval Plugin]] enables users to connect [[ChatGPT]] to an instance of a [[vector database]], allowing any information stored in the connected database to be used to answer questions and provide responses based on the details stored in the database. Additionally, the vector database can be utilized as a long-term storage solution for ChatGPT, allowing it to persist and store portions of user conversations beyond the short-lived memory of a browser tab.
-==The ChatGPT Retrieval Plugin==
+===Plugin Functionality===
-The ChatGPT Retrieval Plugin enables users to connect ChatGPT to a vector database. This connection allows the AI to access custom data stored in the database and answer questions based on the provided information. Additionally, the vector database can serve as a long-term storage system for ChatGPT's memory, enabling it to retain and recall parts of users' conversations.
+The main functions of the ChatGPT Retrieval Plugin include:
+*Connecting a vector database with proprietary data to ChatGPT, allowing it to answer specific questions based on that data
+*Persisting personal documents and details to provide a personalized touch to ChatGPT's responses
+*Storing conversations with ChatGPT in the attached vector database, enabling continued conversations across multiple sessions
-The retrieval plugin can be used to accomplish tasks grounded in the context of knowledge stored in a connected vector database like Weaviate. This process involves two steps: first, ChatGPT is prompted with a question, which it uses to craft a query that is sent to the vector database; second, the vector database provides relevant information to ChatGPT, which it uses to formulate an answer to the user's prompt.
+This functionality allows for regular updates to content stored in connected vector databases, giving the model awareness of new information without the need for costly and time-consuming retraining of the [[large language model]] ([[LLM]]).
-===Main Features of the Retrieval Plugin===
+==Overview of Vector Database Providers==
-Attach a vector database containing proprietary data for ChatGPT to use when answering specific questions.
+This article presents an overview of various vector database providers supported by the plugin, highlighting their unique features, performance, and pricing. The choice of a vector database provider depends on specific use cases and requirements, and each provider necessitates the use of distinct Dockerfiles and environment variables. Detailed instructions for setting up and using each provider can be found in their respective documentation at /docs/providers/<datastore_name>/setup.md.
-Persist personal documents and details, enabling ChatGPT to provide personalized responses based on the user's unique context.
-Store conversations with ChatGPT in the attached vector database to maintain continuity between sessions.
+===Pinecone===
-With these capabilities, users can schedule regular updates to the content stored in the connected vector database, giving the model awareness of new information and reducing the need for frequent, costly retraining of the LLM model.
+[[Pinecone]] is a managed vector database engineered for rapid deployment, speed, and scalability. It uniquely supports hybrid search and is the sole datastore that natively accommodates SPLADE sparse vectors. For comprehensive setup guidance, refer to /docs/providers/pinecone/setup.md.
+===Weaviate===
+[[Weaviate]] is an open-source vector search engine designed to scale effortlessly to billions of data objects. Its out-of-the-box support for hybrid search makes it ideal for users who need efficient keyword searches. Weaviate can be self-hosted or managed, offering flexible deployment options. For extensive setup guidance, refer to /docs/providers/weaviate/setup.md.
+===Zilliz===
+[[Zilliz]] is a managed, cloud-native vector database tailored for billion-scale data. It boasts a plethora of features, including numerous indexing algorithms, distance metrics, scalar filtering, time-travel searches, rollback with snapshots, full RBAC, 99.9% uptime, separated storage and compute, and multi-language SDKs. For comprehensive setup guidance, refer to /docs/providers/zilliz/setup.md.
+===Milvus===
+[[Milvus]] is an open-source, cloud-native vector database that scales to billions of vectors. As the open-source variant of Zilliz, Milvus shares many features with it, such as various indexing algorithms, distance metrics, scalar filtering, time-travel searches, rollback with snapshots, multi-language SDKs, storage and compute separation, and cloud scalability. For extensive setup guidance, refer to /docs/providers/milvus/setup.md.
+===Qdrant===
+[[Qdrant]] is a vector database that can store documents and vector embeddings. It provides self-hosted and managed Qdrant Cloud deployment options, catering to users with diverse requirements. For comprehensive setup guidance, refer to /docs/providers/qdrant/setup.md.
+===Redis===
+[[Redis]] is a real-time data platform suitable for an array of applications, including AI/ML workloads and everyday use. By creating a Redis database with the Redis Stack docker container, Redis can be employed as a low-latency vector engine. For a hosted or managed solution, Redis Cloud is available. For extensive setup guidance, refer to /docs/providers/redis/setup.md.
+===LlamaIndex===
+[[LlamaIndex]] serves as a central interface to connect your LLMs with external data. It offers a collection of in-memory indices over structured and unstructured data for use with ChatGPT. Unlike conventional vector databases, LlamaIndex supports a wide array of indexing strategies (e.g., tree, keyword table, knowledge graph) optimized for various use-cases. It is lightweight, user-friendly, and requires no additional deployment. Users need only specify a few environment variables and, optionally, point to an existing saved Index JSON file. However, metadata filters in queries are not yet supported. For comprehensive setup guidance, refer to /docs/providers/llama/setup.md.
 ==Weaviate Retrieval Plugin in Action==
@@ Line 25: / Line 49: @@
 ===Using ChatGPT on Proprietary Company Documents===
-The Weaviate Retrieval Plugin can be used to create an HR chatbot by training a customized version of ChatGPT on a company's internal documents. This HR chatbot can then assist employees with questions related to onboarding processes, signing up for health benefits, and more
+The Weaviate Retrieval Plugin can be used to create a customized version of ChatGPT trained on a company's internal documents, enabling it to act as a human resources chatbot. This can provide employees with easy access to information about onboarding processes, health benefits, and more.
 ===Personalizing ChatGPT===
-The Weaviate Retrieval Plugin also allows users to customize ChatGPT based on their personal details. By adding information about friends, preferences, and personal characteristics into Weaviate, ChatGPT can provide tailored recommendations, such as planning a menu for an upcoming party or suggesting suitable activities based on the user's interests and abilities.
+The plugin also enables the customization of ChatGPT around personal details, such as information about friends or languages spoken by the user. By storing these details in Weaviate, ChatGPT can provide more tailored and personalized responses.
 ===Helping ChatGPT Remember===
-One of the most exciting applications of the Weaviate Retrieval Plugin is its ability to persist the memory of previous conversations with ChatGPT. By storing chat content in Weaviate, users can ensure that the AI remembers its past interactions, even when sessions are closed and restarted. This feature greatly enhances the continuity and relevance of ChatGPT's responses, making it more akin to a human conversational partner.
+One of the most powerful applications of the Weaviate Retrieval Plugin is its ability to store and reference previous conversations with ChatGPT. By persisting these conversations in Weaviate, ChatGPT can recall past interactions and provide more contextually relevant responses.
-In summary, the ChatGPT Retrieval Plugin, particularly when used with Weaviate, significantly expands the capabilities of ChatGPT by providing access to custom data and long-term memory storage. This plugin enables users to create personalized and context-aware AI chatbots for a wide range of applications, from corporate use to individual customization.
+[[Category:Plugins]] [[Category:ChatGPT Plugins]]