Reflecting on a Year of Innovation

As we look back on 2024, it's clear that Graphlit has experienced a year of remarkable growth and innovation. Our commitment to enhancing developer workflows, improving user experiences, and integrating cutting-edge technologies has driven numerous updates across our platform. From January through December, we've introduced new features, significant enhancements, and critical bug fixes to ensure our SaaS product remains a leader in the industry. Here's a comprehensive reflection on our journey through the year.

January 2024

The year kicked off with substantial advancements in content management and integration capabilities. On January 18th, we introduced support for content publishing, enabling the summarization and repurposing of documents, audio transcripts, and image descriptions. This feature allows developers to efficiently publish conversations as content, generate bulk summaries, and integrate with external services through callback webhooks for LLM tools.

Shortly after, on January 22nd, we expanded our email integration by supporting Google and Microsoft email feeds. This enhancement allows ingestion of both past and new emails, automatically extracting attachments and linking them to parent emails. Additionally, we introduced the reingest content in-place feature, offering more control over content updates and workflow management.

These updates significantly improve developers' ability to manage and utilize various content types, enhancing the robustness and flexibility of applications built on our platform.

February 2024

February brought enhancements focused on semantic processing and model performance. On February 2nd, we launched Semantic Alerts, enabling periodic LLM summarization and content publishing to generate daily reports from multiple feeds. This feature is particularly beneficial for teams looking to automate reporting and maintain up-to-date insights.

We also supported the latest OpenAI 0125 model versions for GPT-4 and GPT-3.5 Turbo, with plans to extend support to Azure OpenAI services as they become available. Additionally, Slack feeds received an upgrade with the inclusion of a listing type field, allowing developers to specify whether to ingest past or new messages.

Later in the month, on February 21st, we introduced support for OneDrive and Google Drive feeds, allowing seamless ingestion of files from shared drives. This update enhances file management capabilities and simplifies the integration process for applications relying on cloud storage solutions. We also added the ability to extract images from PDFs automatically, linking them as children of their parent files, which improves the handling of multimedia content.

These enhancements empower developers with more efficient content workflows and better integration with popular cloud services, ensuring smoother operations and enhanced application functionality.

March 2024

March was marked by significant expansions in model support and telemetry. On March 10th, we unveiled support for Claude 3, Mistral, and Groq models, broadening the range of AI models developers can leverage. Alongside this, we introduced usage and credits telemetry, providing detailed insights into API usage and credit consumption, which helps in monitoring and optimizing resource allocation.

Mid-March, on March 13th, we added support for the Claude 3 Haiku model and enabled direct ingestion of Base64 encoded files, simplifying the process of handling various file formats and improving data ingestion workflows. Furthermore, March 23rd saw the addition of support for Linear, GitHub Issues, and Jira issue feeds, allowing seamless ingestion of issue tracking data as searchable content items. This integration facilitates better project management and tracking within developer workflows.

These updates significantly enhance the platform's capabilities, offering developers more options for model usage and better integration with project management tools, thereby improving overall productivity and efficiency.

April 2024

April was a transformative month with the introduction of robust SDK support and further model integrations. On April 7th, we launched support for Discord feeds, enabling ingestion of messages and file attachments from Discord channels using bot tokens. Additionally, we introduced Cohere reranking, allowing semantic search results to be reordered for increased relevance within RAG pipelines.

Continuing this momentum, on April 23rd, we introduced native Python and TypeScript SDKs. These SDKs are built using Pydantic and TypeScript types and are automatically generated from our GraphQL schema, simplifying the development process by eliminating the need for GraphQL knowledge. Moreover, we expanded support for the latest OpenAI, Cohere, and Groq models, enhancing the versatility and performance of our platform.

These enhancements streamline the development process, providing developers with powerful tools and improved model support, thereby facilitating the creation of more sophisticated and efficient applications.

May 2024

May focused on enhancing search capabilities and improving performance. On May 5th, we introduced support for Jina and Pongo rerankers, offering developers additional options for semantic search result prioritization. Additionally, we added support for Microsoft Teams feeds, enabling the ingestion of messages from Teams channels, which enhances collaboration and communication integration.

On May 15th, we rolled out support for GraphRAG and the OpenAI GPT-4o model. These updates include LLM revisions in RAG conversations, which allow the LLM to refine initial responses, resulting in higher quality outputs and increased token usage by 35%. We also introduced graph visualization in promptConversation responses, providing a visual representation of entity relationships, which aids in better data comprehension and analysis.

These updates offer developers improved search functionality and enhanced data visualization capabilities, leading to more accurate and insightful application performance.

June 2024

June was characterized by significant advancements in model support and semantic search. On June 9th, we added support for Deepseek models for prompt completion and enabled the parsing of embedded JSON-LD from web pages, automatically injecting them into the knowledge graph. Additionally, we enhanced our knowledge graph generation by making it opt-in, allowing developers to control its inclusion in responses more effectively.

On June 21st, we introduced support for the Anthropic Claude 3.5 Sonnet model and enabled semantic search for observable entities within the knowledge graph using vector embeddings. This allows for more accurate and contextually relevant search results, improving the overall user experience and data retrieval processes.

These enhancements provide developers with advanced model capabilities and more precise search functionalities, fostering the development of smarter and more responsive applications.

July 2024

July brought a series of essential updates focusing on model expansions and workflow improvements. On July 4th, we introduced support for webhook Alerts, enabling HTTP POST notifications with published text results, enhancing real-time alert management. We also upgraded Deepseek models to support a 128k token context window, significantly increasing the capacity for handling large datasets and complex queries.

Later in the month, on July 19th and 25th, we expanded our model support to include the OpenAI GPT-4o Mini and Mistral Large 2 models, respectively, offering developers even more options for deploying and managing AI models within their applications. Additionally, on July 28th, we added support for the indexing workflow stage and Azure AI language detection, enhancing our content indexing and language identification capabilities.

These updates enhance the platform's flexibility and performance, providing developers with more powerful tools and greater control over their AI integrations and workflows.

August 2024

August was a pivotal month with significant upgrades in document extraction and entity support. On August 8th, we rolled out support for LLM-based document preparation using models like OpenAI GPT-4o and Anthropic Sonnet 3.5, and introduced an open-source .NET SDK available on Nuget.org with comprehensive code samples on GitHub. This addition makes it easier for developers working with .NET to integrate Graphlit's capabilities into their applications.

On August 11th, we enhanced our platform by supporting Azure AI Document Intelligence by default, improving fidelity in complex PDFs and enabling better table extraction. We also introduced language-aware summaries and entity descriptions, ensuring that summaries and entity metadata align with the source text's language, thereby enhancing the accuracy and relevance of content processing.

Finally, on August 20th, we added support for medical-related entities and Anthropic prompt caching, expanding our platform's capabilities to handle specialized datasets and improving performance by reducing token costs during prompt processing.

These enhancements provide developers with robust document processing capabilities and specialized entity support, facilitating the creation of more precise and efficient applications.

September 2024

September focused on advanced AI integrations and medical data handling. On September 1st, we introduced support for FHIR enrichment, allowing entity enrichment from FHIR servers for medical-related entities, which is essential for healthcare applications. We also added support for the latest Cohere models and the Azure AI Document Intelligence v4.0 preview API, improving model performance and document processing.

On September 26th and 30th, we expanded our model support to include the latest Google Gemini models and Azure AI Inference models, providing developers with access to a broader range of AI capabilities and improving the flexibility of model integrations. These updates ensure that Graphlit remains at the forefront of AI advancements, offering developers the tools they need to build sophisticated and intelligent applications.

These updates significantly enhance the platform's AI capabilities and specialized data handling, making it easier for developers to integrate advanced AI features and manage complex datasets effectively.

October 2024

October saw a blend of tool calling enhancements and model support expansions. On October 3rd, we introduced the ingestBatch mutation, allowing asynchronous ingestion of multiple URIs into content objects, streamlining the content ingestion process. Additionally, support for Google Gemini Flash 1.5 8b models was added, enhancing the range of available AI models.

On October 7th and 22nd, we expanded our tool calling support to include Anthropic and Gemini models, improving the flexibility and integration of external tools within conversations. We also updated Cohere image embeddings, offering developers more robust image processing capabilities.

Furthermore, on October 21st, we introduced support for configuring image and text embedding models at the project level, including models like OpenAI Embedding-3-Small, Cohere Embed 3.0, and Jina CLIP image embeddings. These updates provide developers with greater control over embedding configurations, allowing for more tailored and effective data processing.

Finally, on October 22nd, we added support for the latest Anthropic Sonnet 3.5 model and Cohere image embeddings, ensuring that developers have access to the most recent advancements in AI technology.

These enhancements offer developers improved integration capabilities and access to the latest AI models, enabling the creation of more powerful and flexible applications.

November 2024

November was a month of advanced image analysis and multi-turn summarization. On November 4th, we supported the Anthropic Claude 3.5 Haiku model, enhancing our AI model offerings. We also introduced an automatic disabling feature for all feeds upon reaching the free tier quota, giving developers better control over their usage limits.

On November 10th, we added support for web search mutations, enabling the selection of search services like Tavily or Exa.AI without ingesting entire web pages. Additionally, we launched multi-turn content summarization and enhanced Deepgram language detection, providing more nuanced and accurate summaries and language processing capabilities.

Later in the month, on November 16th and 24th, we introduced multi-turn text summarization and multi-turn image analysis capabilities. These features allow developers to engage in more interactive and detailed conversations with LLMs, facilitating deeper analysis and more refined outputs.

These updates enhance the platform's capability to handle complex interactions and multimedia content, providing developers with more powerful tools for content analysis and summarization.

December 2024

As the year draws to a close, December introduced several pivotal updates that enhanced AI model support, document management, and content publishing capabilities. On December 1st, we implemented support for a retrieval-only RAG pipeline, allowing developers to format LLM-ready prompts using the `formatConversation` and `completeConversation` mutations. This feature streamlines data retrieval processes and enhances the interaction between LLMs and retrieved content.

Continuing on December 10th, we expanded support for OpenAI GPT-4 Turbo 128k, alongside Llama 2 and Mistral models, providing developers with access to some of the most advanced AI models available. Additionally, we introduced query by example for contents and conversations, utilizing vector embeddings to improve search relevancy and user experience.

On December 9th and 27th, we focused on website mapping and Google Docs integration. The December 9th update enabled website mapping using `mapWebmutation` to retrieve URLs from sitemap.xml files, and introduced the ability to generate web page screenshots with optional image processing workflows via `screenshotPagemutation`. On December 27th, we enhanced support for LLM fallbacks to safeguard against model provider downtimes and improved native Google Docs formats ingestion, allowing seamless auto-exportation to Microsoft Office formats.

Finally, on December 22nd, we broadened our integration capabilities by adding support for Dropbox, Box, Intercom, and Zendesk feeds, enabling efficient ingestion from these popular cloud services. We also introduced support for the OpenAI o1 and Gemini 2.0 models, along with various bug fixes to enhance platform stability and performance.

These comprehensive updates showcase Graphlit's dedication to providing developers with versatile tools and robust integrations, ensuring that applications built on our platform are both powerful and reliable.

Looking Ahead

As we reflect on the past year, we're proud of the strides we've made in enhancing the Graphlit platform. Each update has been driven by our commitment to providing developers with the tools they need to build innovative and reliable applications. We're grateful for the support and feedback from our user community, which has been instrumental in guiding our development efforts.

Looking forward, we are excited about the upcoming innovations and improvements slated for the next year. Stay tuned for more advanced features, enhanced integrations, and continued performance optimizations as we strive to make Graphlit an even more powerful ally in your development journey.

Thank you for being a part of our community and for driving the future of Graphlit with us!

Using Graphlit to Publish this Year in Review

For this year in review, it was automatically generated using Graphlit's robust publishing feature, by summarizing each changelog page, and using OpenAI o1-mini to publish the final article.

You can try this out for yourself with this Colab notebook.

Summary

Please email any questions on this article or the Graphlit Platform to questions@graphlit.com.

For more information, you can read our Graphlit Documentation, visit our marketing site, or join our Discord community.