Developer data platform company Couchbase has been busy. It’s flagship product Capella now features new AI services designed to help organisations create intelligent applications with agentic AI functions. With AI agent technologies now capable of handling more complex workflow tasks alongside humans (often with little or no human-in-the-loop intervention) Couchbase is aiming to simplify AI creation workflows and make more AI models available in one unified zone.
Couchbase is perhaps echoing moves seen at Amazon Web Services. As part of its re:Invent 2024 event newsfeed, AWS explained how it is broadening the Amazon Bedrock managed AI application service layer beyond its initial promise to help developers access a variety of foundation models to build AI apps. Amazon Bedrock has now been extended to enable developers to work across models including Mistral, Meta’s Llama, stability.ai and Anthropic.
Model hosting
Aiming to provide model diversity in what feels like a similar vein, Couchbase says its AI Services include model hosting, automated vectorisation (a compiler optimisation technique where the compiler automatically converts scalar code processing one data element at a time into vectorized code), unstructured data preprocessing and AI agent catalogue services. It’s a smorgasbord of AI tooling that is meant to allow organisations to prototype, build, test and deploy AI agents while keeping models and data close together.
This allows AI development teams to avoid the excess latencies and high operational costs that developers can experience when introducing new technology components and workflows. Couchbase says it helps teams to safely bring agent-based applications into production by giving developers control over data across the development lifecycle and mitigating security and privacy issues from large language models (LLMs) running outside the organisation.
“As AI transforms applications, organisations need secure ways to integrate it and must handle the surge of data across diverse formats to build and deploy agentic applications with confidence,” said Matt McDonough, SVP of product and partners at Couchbase. “Couchbase is making this possible by providing a comprehensive AI-powered developer data platform that streamlines retrieval-augmented generation (RAG) pipelines, ensures fast and secure model interactions and enables agent reuse during development and production. We’re helping customers through the broad spectrum of AI advances, from simple vector search to RAG chatbots and sophisticated agentic AI apps, while enabling them to deliver more personalized and contextual experiences.”
Agent-LLM interaction management
Carl Olofson, research VP at IDC thinks that the proliferation of AI agents is fundamentally transforming both development and operations, creating new data management challenges for enterprises. He reminds us that organisations must now preserve and analyse vast amounts of data from agent-LLM interactions, including prompts, responses and validation artefacts – all critical for ensuring ongoing accuracy and reliability.
“Organisations also face the complexity of maintaining agent guardrails against evolving LLM behaviours,” said Olofson. “Couchbase Capella and its new AI Services are designed to strategically address these challenges and provide enterprises the scalable architecture and flexibility needed to handle complex RAG workflows and manage huge volumes of new types of AI data.”
Enterprises adopting AI face growing concerns over data security and privacy, performance and latencies, control and costs and a host of challenges that hinder progress. These include managing different data types and complex data integration workflows, addressing LLM security and accuracy risks, ensuring fast application response times and adapting to rapidly evolving AI tools and data platforms.
Couchbase says Capella AI Services addresses these challenges to help organisations boost developer velocity with model service, which offers managed endpoints for leading LLMs and embedding models. It also provides value-added capabilities, such as prompt and conversation caching, guardrails and keyword filtering to support RAG and agentic workflows. With data and AI models within the same organisational boundary, enterprises can more easily meet privacy and latency requirements for enterprise-grade RAG applications, eliminating the need for costly private links or custom solutions.
“Developer productivity and empowering the development of AI-driven applications anywhere are at the heart of our strategy,” continued McDonough. “By providing a unified platform that supports development from prototype to production, we’re enabling organizations to build and deploy AI applications from cloud-to-edge that are sophisticated, trustworthy and cost effective.”
A menu of AI services
Unstructured Data Services: Extracts, cleans, chunks and transforms unstructured documents into JSON, preparing them for vectorisation. It also extracts structured information from complex documents and makes them queryable in Capella. This saves developers time associated with building DIY preprocessing pipelines.
Vectorisation Services: Automates vectorisation and indexing of data stored in Capella. Along with Capella Model and Unstructured Data Services, this helps developers build a RAG pipeline with fewer tools.
AI Agent Catalog Services: Accelerates agentic application development by offering a centralized repository for tools, metadata, prompts and audit information for LLM flow, traceability and governance. It also automates the discovery of relevant agent tools to answer user questions and manages guardrails to ensure that agent exchanges are consistent over time.
Capella AI Functions: Enhances developer productivity by enabling AI-driven data analysis directly into application workflows using familiar SQL++ syntax. This accelerates developer productivity by eliminating the need for external tooling, custom coding and managing model deployments. Capella AI Functions include summarization, classification, sentiment analysis and data masking.
Developers require a unified platform to build and manage the emerging class of agent-based applications and need specialised data management tools to handle complex AI workflows and language model interactions effectively and securely. McDonough and team say that with Capella AI Services now part of Couchbase’s robust developer data platform, users can streamline RAG pipelines, collect, evaluate and validate agent exchanges and accelerate time to market for RAG and agentic AI applications.
Streamlining complex AI workflows
“Couchbase Capella enables developers to accelerate the creation of agent-based AI applications using their existing experience, skills and preferred tools, including SQL++ query language and over a dozen programming languages,” said McDonough. “Capella streamlines complex AI workflows – from data vectorization and storage to model interactions, validation and analytics. Developers can use automated processing, integrate with popular AI models and frameworks and deploy across multiple clouds, all while maintaining complete control over their data security, LLM interactions, agent exchanges, application performance and the user experiences they create.”
This flexible approach is hoped to allow teams to build sophisticated AI applications without generating excessive expenses, compromising on their specific requirements or having to learn entirely new toolsets. Capella AI Services are available in private preview.