Introduction OpenAI announced the Responses API, their most advanced and versatile interface for building intelligent AI applications. Supporting both text and image inputs with rich text outputs, this API enables dynamic, stateful conversations that remember and build on previous interactions, making AI experiences more natural and context-aware. It also unlocks powerful capabilities through built-in tools such as web search, file search, code interpreter, and more, while enabling seamless integration with external systems via function calling. Its event-driven design delivers clear, structured updates at every step, making it easier than ever to create sophisticated, multi-step AI workflows. Key features include: Stateful conversations via the previous response ID Built-in tools like web search, file search, code interpreter, MCP, and others Access to advanced models available exclusively, such as o1-pro Enhanced support for reasoning models with reasoning summaries and efficient context management through previous response ID or encrypted reasoning items Clear, event-based outputs that simplify integration and control While the Chat Completions API remains fully supported and widely used, OpenAI plans to retire the Assistants API in the first half of 2026. To support the adoption of the Responses API, two new Snaps have been introduced: OpenAI Chat Completions ⇒ OpenAI Responses API Generation OpenAI Tool Calling ⇒ OpenAI Responses API Tool Calling Both Snaps are fully compatible with existing upstream and downstream utility Snaps, including the OpenAI Prompt Generator, OpenAI Multimodal Content Generator, all Function Generators (Multi-Pipeline, OpenAPI, and APIM), the Function Result Generator, and the Message Appender. This allows existing pipelines and familiar development patterns to be reused while gaining access to the advanced features of the Responses API. OpenAI Responses API Generation   The OpenAI Responses API Generation Snap is designed to support OpenAI’s newest Responses API, enabling more structured, stateful, and tool-augmented interactions. While it builds upon the familiar interface of the Chat Completions Snap, several new properties and behavioral updates have been introduced to align with the Responses API’s capabilities. New properties Message: The input sent to the LLM. This field replaces the previous Use message payload, Message payload, and Prompt properties in the OpenAI Chat Completions Snap, consolidating them into a single input. It removes ambiguity between "prompt" as raw text and as a template, and supports both string and list formats. Previous response ID: The unique ID of the previous response to the model. Use this to create multi-turn conversations. Model parameters Reasoning summary: For reasoning models, provides a summary of the model’s reasoning process, aiding in debugging and understanding the model's reasoning process. The property can be none, auto, or detailed. Advanced prompt configurations Instructions: Applied only to the current response, making them useful for dynamically swapping instructions between turns. To persist instructions across turns when using previous_response_id, the developer message in the OpenAI Prompt Generator Snap should be used. Advanced response configurations Truncation: Defines how to handle input that exceeds the model’s context window. auto allows the model to truncate the middle of the conversation to fit, while disabled (default) causes the request to fail with a 400 error if the context limit is exceeded. Include reasoning encrypted content: Includes an encrypted version of reasoning tokens in the output, allowing reasoning items to persist when the store is disabled. Built-in tools Web search: Enables the model to access up-to-date information from the internet to answer queries beyond its training data. Web search type Search context size User location: an approximate user location including city, region, country, and timezone to deliver more relevant search results. File search: Allows the model to retrieve information from documents or files. Vector store IDs Maximum number of results Include search results: Determines whether raw search results are included in the response for transparency or debugging. Ranker Score threshold  Filters: Additional metadata-based filters to refine search results. For more details on using filters, see Metadata Filtering. Advanced tool configuration Tool choice: A new option, SPECIFY A BUILT-IN TOOL, allows specifying that the model should use a built-in tool to generate a response. Note that the OpenAI Responses API Generation Snap does not support the response count or stop sequences properties, as these are not available in the Responses API. Additionally, the message user name, which may be specified in the Prompt Generator Snap, is not supported and will be ignored if included. Model response of Chat Completions vs Responses API   Chat Completions API Responses API   The Responses API introduces an event-driven output structure that significantly enhances how developers build and manage AI-powered applications compared to the traditional Chat Completions API. While the Chat Completions API returns a single, plain-text response within the choices array, the Responses API provides an output array containing a sequence of semantic event items—such as reasoning, message, function_call, web_search_call, and more—that clearly delineate each step in the model's reasoning and actions. This structured approach allows developers to easily track and interpret the model's behavior, facilitating more robust error handling and smoother integration with external tools. Moreover, the response from the Responses API includes the model parameters settings, providing additional context for developers. Pipeline examples Built-in tool: web search This example demonstrates how to use the built-in web search tool. In this pipeline, the user’s location is specified to ensure the web search targets relevant geographic results. System prompt: You are a friendly and helpful assistant. Please use your judge to decide whether to use the appropriate tools or not to answer questions from the user. Prompt: Can you recommend 2 good sushi restaurants near me? Output: As a result, the output contains both a web search call and a message. The model uses the web search to find and provide recommendations based on current data, tailored to the specified location. Built-in tool: File search This example demonstrates how the built-in file search tool enables the model to retrieve information from documents stored in a vector store during response generation. In this case, the file wildfire_stats.pdf has been uploaded. You can create and manage vector stores through the Vector Store management page. Prompt: What is the number of Federal wildfires in 2018 Output: The output array contains a file_search_call event, which includes search results in its results field. These results provide matched text, metadata, and relevance scores from the vector store. This is followed by a message event, where the model uses the retrieved information to generate a grounded response. The presence of detailed results in the file_search_call is enabled by selecting the Include file search results option. OpenAI Responses API Tool Calling               The OpenAI Responses API Tool Calling Snap is designed to support function calling using OpenAI’s Responses API. It works similarly to the OpenAI Tool Calling Snap (which uses the Chat Completions API), but is adapted to the event-driven response structure of the Responses API and supports stateful interactions via the previous response ID. While it shares much of its configuration with the Responses API Generation Snap, it is purpose-built for workflows involving function calls. Existing LLM agent pipeline patterns and utility Snaps—such as the Function Generator and Function Result Generator—can continue to be used with this Snap, just as with the original OpenAI Tool Calling Snap. The primary difference lies in adapting the Snap configuration to accommodate the Responses API’s event-driven output, particularly the structured function_call event item in the output array. The Responses API Tool Calling Snap provides two output views, similar to the OpenAI Tool Calling Snap, with enhancements to simplify building agent pipelines and support stateful interactions using the previous response ID: Model response view: The complete API response, including extra fields: messages: an empty list if store is enabled, or the full message history—including messages payload and model response—if disabled (similar to the OpenAI Tool Calling Snap). When using stateful workflows, message history isn’t needed because the previous response ID is used to maintain context. has_tool_call: a boolean indicating whether the response includes a tool call. Since the Responses API no longer includes the  finish_reason: "tool_calls"  field, this new field makes it easier to create stop conditions in the pipeloop Snap within the agent driver pipeline. Tool call view: Displays the list of function calls made by the model during the interaction. Tool Call View of Chat Completions vs Responses API   Uses id as the function call identifier when sending back the function result. Tool call properties (name, arguments) are nested inside the function field.   Each tool call includes: • id: the unique event ID • call_id: used to reference the function call when returning the result The tool call structure is flat — name and arguments are top-level fields. Building LLM Agent Pipelines To build LLM agent pipelines with the OpenAI Responses API Tool Calling Snap, you can reuse the same agent pipeline pattern described in Introducing Tool Calling Snaps and LLM Agent Pipelines. Only minor configuration changes are needed to support the Responses API. Agent Driver Pipeline The primary change is in the PipeLoop Snap configuration, where the stop condition should now check the has_tool_call field, since the Responses API no longer includes the finish_reason:"tool_calls". Agent Worker Pipeline Fields mapping A Mapper Snap is used to prepare the related fields for the OpenAI Responses API Tool Calling Snap. OpenAI Responses API Tool Calling The key changes are in this Snap’s configuration to support the Responses API’s stateful interactions. There are two supported approaches: Option 1: Use Store (Recommended) Leverages the built-in state management of the Responses API. Enable Store Use Previous Response ID Send only the function call results as the input messages for the next round. (messages field in the Snap’s output will be an empty array, so you can still use it in the Message Appender Snap to collect tool results.) Option 2: Maintain Conversation History in Pipeline Similar to the approach used in the Chat Completions API. Disable Store Include the full message history in the input (messages field in the Snap’s output contains message history) (Optional) Enable Include Reasoning Encrypted Content (for reasoning models) to preserve reasoning context efficiently OpenAI Function Result Generator As explained in Tool Call View of Chat Completions vs Responses API section, the Responses API includes both an id and a call_id. You must use the call_id to construct the function call result when sending it back to the model. Conclusion The OpenAI Responses API makes AI workflows smarter and more adaptable, with stateful interactions and built-in tools. SnapLogic’s OpenAI Responses API Generation and Tool Calling Snaps bring these capabilities directly into your pipelines, letting you take advantage of advanced features like built-in tools and event-based outputs with only minimal adjustments. By integrating these Snaps, you can seamlessly enhance your workflows and fully unlock the potential of the Responses API.

    This document explores the integration of Google Cloud's Vertex AI Retrieval Augmented Generation (RAG) capabilities with SnapLogic. We will delve into how Vertex AI RAG functions, its benefits over traditional vector databases, and practical applications within the SnapLogic platform. The guide will cover setting up and utilizing Vertex AI RAG, automating knowledge feeds, and integrating with SnapLogic's Generate snaps for enhanced LLM performance. Vertex AI RAG Engine The Vertex AI RAG Engine streamlines the retrieval-augmented generation (RAG) process through two primary steps: Knowledge Management: The Vertex AI RAG Engine establishes and maintains a knowledge base by creating a corpus, which serves as an index for storing source files. Retrieval Query: Upon receiving a prompt, the Vertex AI RAG Engine efficiently searches this knowledge base to identify and retrieve information most relevant to the request. The Vertex AI RAG Engine integrates Google Cloud's Vertex AI with the RAG architecture to produce accurate and contextually relevant LLM responses. It covers tasks related to managing knowledge by creating a corpus as an index for source files. For processing, it efficiently retrieves relevant information from this knowledge base when a prompt is received, then leverages the LLM to generate a response based on the retrieved context. Difference between Vector Database While both traditional vector databases and the Vertex AI RAG Engine are designed to enhance LLM responses by providing external knowledge, they differ significantly in their approach and capabilities. Vector Databases Vector databases primarily focus on storing and querying vector embeddings. To use them with an LLM for RAG, you typically need to: Manually manage embedding generation: You are responsible for generating vector embeddings for your source data using an embedding model. Handle retrieval logic: You need to implement the logic for querying the vector database, retrieving relevant embeddings, and then mapping them back to the original source text. Integrate with LLM: The retrieved text then needs to be explicitly passed to the LLM as part of the prompt. No built-in LLM integration: They are agnostic to the LLM and require manual integration for RAG workflows. Vertex AI RAG Engine The Vertex AI RAG Engine offers a more integrated and streamlined solution, abstracting away much of the complexity. Key differences include: Integrated knowledge management: It handles the entire lifecycle of your knowledge base, from ingesting raw source files to indexing and managing the corpus. You don't need to manually generate embeddings or manage vector storage. Automated retrieval: The engine automatically performs the retrieval of relevant information from its corpus based on the user's prompt. Seamless LLM integration: It's designed to work directly with Vertex AI's LLMs, handling the contextualization of the prompt with retrieved information before passing it to the LLM. End-to-end solution: It provides a more comprehensive solution for RAG, simplifying the development and deployment of RAG-powered applications. In essence, a traditional vector database is a component that requires significant orchestration to implement RAG. In contrast, the Vertex AI RAG Engine is a more complete, managed service that simplifies the entire RAG workflow by providing integrated knowledge management, retrieval, and LLM integration. This fundamental benefit allows for a significant simplification of the often complex RAG processing pipeline. By streamlining this process, we can achieve greater efficiency, reduce potential points of failure, and ultimately deliver more accurate and relevant results when leveraging large language models (LLMs) for tasks that require external knowledge. This simplification not only improves performance but also enhances the overall manageability and scalability of RAG-based systems, making them more accessible and effective for a wider range of applications. Using Vertex AI's RAG Engine with Generative AI (instead of directly via the Gemini API) offers advantages. It enhances query-related information retrieval through built-in tools, streamlining data access for generative AI models. This native integration within Vertex AI optimizes information flow, reduces complexity, and leads to a more robust system for retrieval-augmented generation.   Vertex AI RAG Engine in SnapLogic SnapLogic now includes a set of Snaps for utilizing the Vertex AI RAG Engine. Corpus Management The following Snaps are available for managing RAG corpora: Google Vertex AI RAG Create Corpus Google Vertex AI RAG List Corpus Google Vertex AI RAG Get Corpus Google Vertex AI RAG Delete Corpus   File Management in Corpus The following Snaps enable file management within a RAG corpus: Google Vertex AI RAG Corpus Add File Google Vertex AI RAG Corpus List File Google Vertex AI RAG Corpus Get File Google Vertex AI RAG Corpus Remove File Retrieval For performing retrieval operations, use the following Snap: Google Vertex AI RAG Retrieval Query Try using Vertex AI RAG Let's walk through a practical example of how to leverage the Vertex AI RAG Engine within SnapLogic. This section will guide you through setting up a corpus, adding files, performing retrieval queries, and integrating the results into your LLM applications. Preparing step Before integration, two key steps are required: First, set up a Google Cloud project with enabled APIs, linked billing, and necessary permissions. List of required enabled Google API https://console.cloud.google.com/apis/library/cloudresourcemanager.googleapis.com https://console.cloud.google.com/apis/library/aiplatform.googleapis.com SnapLogic offers two primary methods for connecting to Google Cloud APIs: Service Account (recommended): SnapLogic can utilize an existing Service Account that possesses the necessary permissions. OAuth2: This method requires configuring OAuth2. Access Token: An Access Token is a temporary security credential to access Google Cloud APIs. It requires manual refreshing of the token when it expires. Create the corpus To build the corpus, use the Google Vertex AI RAG Create Corpus Snap. Place the Google Vertex AI RAG Create Corpus Snap. Create Google GenAI Service Account Upload the Service account JSON key file that you obtained from Google Cloud Platform, and then select the project and resource location you want to use. We recommend using the “us-central1” location. Edit the configuration by setting the display name and the Snap execution to "Validate & Execute." Validate the pipeline to obtain the corpus result in the output. If the result is similar to the image above, you now have the corpus ready to add the document. Upload the document To upload documents for Google Vertex AI RAG, integrate SnapLogic using a pipeline connecting the "Google Vertex AI RAG Corpus Add File" and "File Reader" Snaps. The "File Reader" accesses the document, passing its content to the "Google Vertex AI RAG Corpus Add File" Snap, which uploads it to a specified Vertex AI RAG corpus.  Example Download the example document. Example file: Basics of SnapLogic.pdf Configure the File Reader Snap as follows: Configure the Corpus Add File Snap as follows: These steps will add the Basics of SnapLogic.pdf to the corpus in the previous section. If you run the pipeline successfully, the output will appear as follows. Retrieve query This section demonstrates how to use the Google Vertex AI RAG Retrieval Query Snap to fetch relevant information from the corpus. This snap takes a user query and returns the most pertinent documents or text snippets. Example From the existing corpus, we will query the question "What snap types does SnapLogic have?" and configure the snap accordingly. The result will display a list of text chunks related to the question, ordered by score value. The score value is calculated from the similarity or distance between the query and each text chunk. The similarity or distance depends on the vectorDB that you choose. By default, the score is the COSINE_DISTANCE. Generate the result Now that we have successfully retrieved relevant information from our corpus, the next crucial step is to leverage this retrieved context to generate a coherent and accurate response using an LLM. This section will demonstrate how to integrate the results from the Google Vertex AI RAG Retrieval Query Snap with a generative AI model, such as the Google Gemini Generate Snap, to produce a final answer based on the augmented information. Here's an example prompt to use in the prompt generator: The final answer will appear as follows: Additionally, the integration between Vertex AI RAG and SnapLogic provides the significant benefit of cross-model compatibility. This means that the established RAG workflows and data retrieval processes can be seamlessly adapted and utilized with different large language models beyond just Vertex AI, such as open-source models or other commercial LLMs. This flexibility allows organizations to leverage their investment in RAG infrastructure across a diverse ecosystem of AI models, enabling greater adaptability, future-proofing of applications, and the ability to choose the best-suited LLM for specific tasks without rebuilding the entire information retrieval pipeline. This cross-model benefit ensures that the RAG solution remains versatile and valuable, regardless of evolving LLM landscapes. Auto-retrieve query with the Vertex AI built-in tool Using the built-in tool in the Vertex AI Gemini Generate Snap for auto-retrieval significantly simplifies the RAG pipeline. Instead of manually performing a retrieval query and then passing the results to a separate generation step, this integrated approach allows the Gemini model to automatically consult the configured RAG corpus based on the input prompt. This reduces the number of steps and the complexity of the pipeline, as the retrieval and generation processes are seamlessly handled within a single Snap. It ensures that the LLM always has access to the most relevant contextual information from your knowledge base without requiring explicit orchestration, leading to more efficient and accurate content generation. The snap configuration example below demonstrates how to configure the Built-in tools section. Specifically, we select the vertexRagStore type and designate the target corpus. The final answer generated using the auto-retrieval process will be displayed below. The response includes grounding metadata for source tracking, allowing users to trace information origins. This feature enhances transparency, fact-verification, and builds trust in content accuracy and reliability. Users can delve into source material, cross-reference facts, and gain a complete understanding, boosting the system's utility and trustworthiness. Summary This document demonstrates how to integrate Google Cloud's Vertex AI Retrieval Augmented Generation (RAG) capabilities with SnapLogic to enhance LLM workflows. Key takeaways include: Streamlined RAG Process: Vertex AI RAG simplifies knowledge management and retrieval, abstracting away complexities like manual embedding generation and retrieval logic, which are typically required with traditional vector databases. Integrated Solution: Unlike standalone vector databases, Vertex AI RAG offers an end-to-end solution for RAG, handling everything from ingesting raw files to integrating with LLMs. SnapLogic Integration: SnapLogic provides dedicated Snaps for managing Vertex AI RAG corpora (creating, listing, getting, deleting), managing files within corpora (adding, listing, getting, removing), and performing retrieval queries. Practical Application: The guide provided a step-by-step example of setting up a corpus, uploading documents, performing retrieval queries using the Google Vertex AI RAG Retrieval Query Snap, and integrating the results with generative AI models like the Google Gemini Generate Snap for contextually accurate responses. Cross-Model Compatibility: A significant benefit of this integration is the ability to adapt established RAG workflows and data retrieval processes with various LLMs beyond just Vertex AI, including open-source and other commercial models, ensuring flexibility and future-proofing. Automated Retrieval with Built-in Tools: The integration allows for automated retrieval via built-in tools in the Vertex AI Gemini Generate Snap, simplifying the RAG pipeline by handling retrieval and generation seamlessly within a single step. By leveraging Vertex AI RAG with SnapLogic, organizations can simplify the development and deployment of RAG-powered applications, leading to more accurate, contextually relevant, and efficient LLM responses.

At SnapLogic, while building and refining an AI Agent for a large customer in the healthcare industry, we embarked on a journey of holistic performance optimization. We didn't just want to make it faster. We tried to make it better across the board. This journey taught us that significant gains are found by looking at the entire system, from the back-end data sources to the pixels on the user's screen. Here’s our playbook for building a truly high-performing AI agent, backed by real-world metrics. The Foundation: Data and Architecture Before you can tune an engine, you have to build it on a solid chassis. For an AI Agent, that chassis is its core architecture and its relationship with data. Choose the Right Brain for the Job: Not all LLMs are created equal. The "best" model depends entirely on the nature of the tasks your agent needs to perform. A simple agent with one or two tools has very different requirements from a complex agent that needs to reason, plan, and execute dynamic operations. Matching the model to the task complexity is key to balancing cost, speed, and capability. Task Complexity Model Type Characteristics & Best For Simple, Single-Tool Tasks Fast & Cost-Effective Goal: Executing a well-defined task with a limited toolset (e.g., simple data lookups, classification). These models are fast and cheap, perfect for high-volume, low-complexity actions. Multi-Tool Orchestration Balanced Goal: Reliably choosing the correct tool from several options and handling moderately complex user requests. These models offer a great blend of speed, cost, and improved instruction-following for a good user experience. Complex Reasoning & Dynamic Tasks High-Performance / Sophisticated Goal: Handling ambiguous requests that require multi-step reasoning, planning, and advanced tool use like dynamic SQL query generation. These are the most powerful (and expensive) models, essential for tasks where deep understanding and accuracy are critical. Deconstruct Complexity with a Multi-Agent Approach: A single, monolithic agent designed to do everything can become slow and unwieldy. A more advanced approach is to break down a highly complex agent into a team of smaller, specialized agents. This strategy offers two powerful benefits: It enables the use of faster, cheaper models. Each specialized agent has a narrower, more defined task, which often means you can use a less powerful (and faster) LLM for that specific job, reserving your most sophisticated model for the "manager" agent that orchestrates the others. It dramatically increases reusability. These smaller, function-specific agents and their underlying tools are modular. They can be easily repurposed and reused in the next AI Agent you build, accelerating future development cycles. Set the Stage for Success with Data: An AI Agent is only as good as the data it can access. We learned that optimizing data access is a critical first step. This involved: Implementing Dynamic Text-to-SQL: Instead of relying on rigid, pre-defined queries, we empowered the agent to build its own SQL queries dynamically from natural language. This flexibility required a deep initial investment in analyzing and understanding the critical columns and data formats our agent would need from sources like Snowflake. Generating Dedicated Database Views: To support the agent, we generated dedicated views on top of our source tables. This strategy serves two key purposes: it dramatically reduces query times by pre-joining and simplifying complex data, and it allows us to remove sensitive or unnecessary data from the source, ensuring the agent only has access to what it needs. Pre-loading the Schema for Agility: Making the database schema available to the agent is critical for accurate dynamic SQL generation. To optimize this, we pre-load the relevant schemas at startup. This simple step saves precious time on every single query the agent generates, contributing significantly to the overall responsiveness. The Engine: Tuning the Agent’s Logic and Retrieval Our Diagnostic Toolkit: Using AI to Analyze AI Before we could optimize the engine, we needed to know exactly where the friction was. Our diagnostic process followed a two-step approach: High-Level Analysis: We started in the SnapLogic Monitor, which provides a high-level, tabular view of all pipeline executions. This dashboard is the starting point for any performance investigation. As you can see below, it gives a list of all runs, their status, and their total duration. By clicking the Download table button, you can export this summary data as a CSV. This allows for a quick, high-level analysis to spot outliers and trends without immediately diving into verbose log files. AI-Powered Deep Dive: Once we identified a bottleneck from the dashboard—a pipeline that was taking longer than expected—we downloaded the detailed, verbose log files for those specific pipeline runs. We then fed these complex logs into an AI tool of our choice. This "AI analyzing AI" approach helped us instantly pinpoint key issues that would have taken hours to find manually. For example, this process uncovered an unnecessary error loop caused by duplicate JDBC driver versions, which significantly extended the execution time of our Snowflake Snaps. Fixing this single issue was a key factor in the 68% performance improvement we saw when querying our technical knowledge base. With a precise diagnosis in hand, we turned our attention to the agent's "thinking" process. This is where we saw some of our most dramatic performance gains. How We Achieved This: Crafting the Perfect Instructions (System Prompts): We transitioned from generic prompts to highly customized system prompts, optimized for both the specific task and the chosen LLM. A simpler model gets a simpler, more direct prompt, while a sophisticated model can be instructed to "think step-by-step" to improve its reasoning. A Simple Switch for Production Speed: One of the most impactful, low-effort optimizations came from how we use a key development tool: the Record Replay Snap. During the creation and testing of our agent's pipelines, this Snap is invaluable for capturing and replaying data, but it adds about 2.5 seconds of overhead to each execution. For a simple agent run involving a driver, a worker, and one tool, this adds up to 7.5 seconds of unnecessary latency in a production environment. Once our pipelines were successfully tested, we switched these Snaps to "Replay Only" mode. This simple change instantly removed the recording overhead, providing a significant speed boost across all agent interactions. Smarter, Faster Data Retrieval (RAG Optimization): For our Retrieval-Augmented Generation (RAG) tools, we focused on two key levers: Finding the Sweet Spot (k value): We tuned the k value—the number of documents retrieved for context. For our product information retrieval use case, adjusting this value was the key to our 63% speed improvement. It’s the art of getting just enough context for an accurate answer without creating unnecessary work for the LLM. Surgical Precision with Metadata: Instead of always performing a broad vector search, we enabled the agent to use metadata. If it knows a document's unique_ID, it can fetch that exact document. This is the difference between browsing a library and using a call number. It's swift and precise. Ensuring Consistency: We set the temperature to a low value during the data extraction and indexing process. This ensures that the data chunks are created consistently, leading to more reliable and repeatable search results. The Results: A Data-Driven Transformation Our optimization efforts led to significant, measurable improvements across several key use cases for the AI Agent. Use Case Before Optimization After Optimization Speed Improvement Querying Technical Knowledge Base 92 seconds 29 seconds ~68% Faster Processing Sales Order Data 32 seconds 10.7 seconds ~66% Faster RAG Retrieval 5.8 seconds 2.1 seconds ~63% Faster Production Optimization (Replay Only) 20 seconds 17.5 seconds ~12% Faster* (*This improvement came from switching development Snaps to a production-ready "Replay Only" mode, removing the latency inherent to the testing phase.) The Experience: Focusing on the User Ultimately, all the back-end optimization in the world is irrelevant if the user experience is poor. The final layer of our strategy was to focus on the front-end application. Engage, Don't Just Wait: A simple "running..." message can cause user anxiety and make any wait feel longer. Our next iteration will provide a real-time status of the agent's thinking process (e.g., "Querying product database...", "Synthesizing answer..."). This transparency keeps the user engaged and builds trust. Guide the User to Success: We learned that a blank text box can be intimidating. By providing predefined example prompts and clearly explaining the agent's capabilities, we guide the user toward successful interactions. Deliver a Clear Result: The final output must be easy to consume. We format our results cleanly, using tables, lists, and clear language to ensure the user can understand and act on the information instantly. By taking this holistic approach, we optimized the foundation, the engine, and the user experience to build an AI Agent that doesn't just feel fast. It feels intelligent, reliable, and genuinely helpful.

Why Security is Essential for Generative AI Applications As generative AI applications transition from prototypes to enterprise-grade solutions, ensuring security becomes non-negotiable. These applications often interact with sensitive user data, internal databases, and decision-making logic that must be protected from unauthorized access. Streamlit, while great for quickly developing interactive AI interfaces, lacks built-in access control mechanisms. Therefore, integrating robust authentication and authorization workflows is critical to safeguarding both the user interface and backend APIs. Overview of the AgentCreator + Streamlit Architecture This guide focuses on securing a generative AI-powered Sales Agent application built with SnapLogic AgentCreator and deployed via Streamlit. The application integrates Salesforce OAuth 2.0 as an identity provider and secures its backend APIs using SnapLogic API Management. Through this setup, only authorized Salesforce users from a trusted domain can access the application, ensuring end-to-end security for both the frontend and backend.   Understanding the Application Stack Role of SnapLogic's AgentCreator Toolkit The SnapLogic AgentCreator Toolkit enables developers and sales engineers to build sophisticated AI-powered agents without having to manage complex infrastructure. These agents operate within SnapLogic pipelines, making it easy to embed business logic, API integrations, and data processing in a modular way. For example, a sales assistant built with AgentCreator and exposed as API using Triggered Tasks can pull real-time CRM data, generate intelligent responses, and return it via a clean web interface. Streamlit as User Interface On the frontend, Streamlit is used to build a simple, interactive web interface for users to query the Sales Agent. Importance of API Management in AI Workflows Once these agents are exposed via HTTP APIs, managing who accesses them—and how—is crucial. That’s where SnapLogic API Management comes in. It provides enterprise-grade tools for API publishing, securing endpoints, enforcing role-based access controls, and monitoring traffic. These features ensure that only verified users and clients can interact with your APIs, reducing the risk of unauthorized data access or abuse. However, the real challenge lies in securing both ends: The Streamlit UI, which needs to restrict access to authorized users. The SnapLogic APIs, exposing the AgentCreator Pipelines which must validate and authorize each incoming request.   OAuth 2.0 Authentication: Fundamentals and Benefits What is OAuth 2.0? OAuth 2.0 is an open standard protocol designed for secure delegated access. Instead of sharing credentials directly, users grant applications access to their resources using access tokens. This model is particularly valuable in enterprise environments, where central identity management is crucial. By using OAuth 2.0, applications can authenticate users through trusted Identity Providers (IDPs) while maintaining a separation of concerns between authentication, authorization, and application logic. Why Use Salesforce as the Identity Provider (IDP)? Salesforce is a robust identity provider that many organizations already rely on for CRM, user management, and security. Leveraging Salesforce for OAuth 2.0 authentication allows developers to tap into a pre-existing user base and organizational trust framework. In this tutorial, Salesforce is used to handle login and token issuance, ensuring that only authorized Salesforce users can access the Streamlit application. This integration also simplifies compliance with enterprise identity policies such as SSO, MFA, and domain-based restrictions. To address the authentication challenge, we use the OAuth 2.0 Authorization Code Flow, with Salesforce acting as both the Identity and Token Provider. Here is Salesforce’s official documentation on OAuth endpoints, which is helpful for configuring your connected app. 🔒 Note: While Salesforce is a logical choice for this example—since the Sales Agent interacts with Salesforce data—any OAuth2-compliant Identity Provider (IDP) such as Google, Okta, or Microsoft Entra ID (formerly Azure AD) can be used. The core authentication flow remains the same, with variations primarily in OAuth endpoints and app registration steps. Architecture Overview and Security Objectives Frontend (Streamlit) vs Backend (SnapLogic APIs) The application architecture separates the frontend interface and backend logic. The frontend is built using Streamlit, which allows users to interact with a visually intuitive dashboard. It handles login, displays AI-generated responses, and captures user inputs. The backend, powered by SnapLogic's AgentCreator, hosts the core business logic within pipelines that are exposed as APIs. This separation ensures flexibility and modular development, but it also introduces the challenge of securing both components independently yet cohesively. Threat Model and Security Goals The primary security threats in such a system include unauthorized access to the UI, data leaks through unsecured APIs, and token misuse. To mitigate these risks, the following security objectives are established: Authentication: Ensure only legitimate users from a trusted identity provider (Salesforce) can log in. Authorization: Grant API access based on user roles and domains, verified via SnapLogic APIM policies. Token Integrity: Validate and inspect access tokens before allowing backend communication with SnapLogic APIM Policies Secret Management: Store sensitive credentials (like Client ID and Secret) securely using Streamlit's secret management features. This layered approach aligns with enterprise security standards and provides a scalable model for future generative AI applications.   Authentication & Authorization Flow Here’s how we securely manage access: 1. Login via Salesforce: Users are redirected to Salesforce’s login screen. After successful login, Salesforce redirects back to the app with an access token. The token and user identity info are stored in Streamlit’s session state.   2. Calling SnapLogic APIs: The frontend sends requests to SnapLogic’s triggered task APIs, attaching the Salesforce access token in the Authorization HTTP Header.   3. Securing APIs via SnapLogic Policies: Callout Authenticator Policy: Validates the token by sending it to Salesforce’s token validation endpoint, as Salesforce tokens are opaque and not self-contained like JWTs. AuthorizeByRole Policy: After extracting the user’s email address, this policy checks if the domain (e.g., @snaplogic.com) is allowed. If so, access is granted.   Below you can find the complete OAuth 2 Authorization Code Flow enhanced with the Token Introspection & Authorization Flow This setup ensures end-to-end security, combining OAuth-based authentication with SnapLogic’s enterprise-grade API Management capabilities. In the following sections, we’ll walk through how to implement each part—from setting up the Salesforce Connected App to configuring policies in SnapLogic—so you can replicate or adapt this pattern for your own generative AI applications. Step 1: Set Up Salesforce Connected App Navigate to Salesforce Developer Console To initiate the OAuth 2.0 authentication flow, you’ll need to register your application as a Connected App in Salesforce. Begin by logging into your Salesforce Developer or Admin account. From the top-right gear icon, navigate to Setup → App Manager. Click on “New Connected App” to create a new OAuth-enabled application profile.   Define OAuth Callback URLs and Scopes In the new Connected App form, set the following fields under the API (Enable OAuth Settings) section: Callback URL: This should be the URL of your Streamlit application (e.g., https://snaplogic-genai-builder.streamlit.app/Sales_Agent). Selected OAuth Scopes: Include at least openid, email, and profile. You may also include additional scopes depending on the level of access required. Ensure that the “Enable OAuth Settings” box is checked to make this app OAuth-compliant.   Retrieve Client ID and Client Secret After saving the app configuration, Salesforce will generate a Consumer Key (Client ID) and a Consumer Secret. These are crucial for the OAuth exchange and must be securely stored. You will use these values later when configuring the Streamlit OAuth integration and environmental settings. Do not expose these secrets in your codebase or version control. 📄 For details on Salesforce OAuth endpoints, see: 👉 Salesforce OAuth Endpoints Documentation   Step 2: Integrate OAuth with Streamlit Using streamlit-oauth Install and Configure streamlit-oauth Package To incorporate OAuth 2.0 authentication into your Streamlit application, you can use the third-party package streamlit-oauth (streamlit-oauth). This package abstracts the OAuth flow and simplifies integration with popular identity providers like Salesforce. To install it, run the following command in your terminal: pip install streamlit-oauth After installation, you'll configure the OAuth2Component to initiate the login process and handle token reception once authentication is successful.   Handle ClientID and ClientSecret Securely Once users log in through Salesforce, the app receives an Access Token and an ID token. These tokens should never be exposed in the UI or logged publicly. Instead, store them securely in st.session_state, Streamlit's native session management system. This ensures the tokens are tied to the user's session and can be accessed for API calls later in the flow.   Store Credentials via Streamlit Secrets Management Storing secrets such as CLIENT_ID and CLIENT_SECRET directly in your source code is a security risk. Streamlit provides a built-in Secrets Management system that allows you to store sensitive information in a .streamlit/secrets.toml file, which should be excluded from version control. Example: # .streamlit/secrets.toml SF_CLIENT_ID = "your_client_id" SF_CLIENT_SECRET = "your_client_secret" In your code, you can access these securely: CLIENT_ID = st.secrets["SF_CLIENT_ID"] CLIENT_SECRET = st.secrets["SF_CLIENT_SECRET"]   Step 3: Manage Environment Settings with python-dotenv Why Environment Variables Matter Managing environment-specific configuration is essential for maintaining secure and scalable applications. In addition to storing sensitive credentials using Streamlit’s secrets management, storing dynamic OAuth parameters such as URLs, scopes, and redirect URIs in an environment file (e.g., .env) allows you to keep code clean and configuration flexible. This is particularly useful if you plan to deploy across multiple environments (development, staging, production) with different settings.   Store OAuth Endpoints in .env Files To manage environment settings, use the python-dotenv package (python-dotenv), which loads environment variables from a .env file into your Python application. First, install the library: pip install python-dotenv Create a .env file in your project directory with the following format: SF_AUTHORIZE_URL=https://login.salesforce.com/services/oauth2/authorize SF_TOKEN_URL=https://login.salesforce.com/services/oauth2/token SF_REVOKE_TOKEN_URL=https://login.salesforce.com/services/oauth2/revoke SF_REDIRECT_URI=https://your-streamlit-app-url SF_SCOPE=id openid email profile Then, use the dotenv_values function to load the variables into your script: from dotenv import dotenv_values env = dotenv_values(".env") AUTHORIZE_URL = env["SF_AUTHORIZE_URL"] TOKEN_URL = env["SF_TOKEN_URL"] REVOKE_TOKEN_URL = env["SF_REVOKE_TOKEN_URL"] REDIRECT_URI = env["SF_REDIRECT_URI"] SCOPE = env["SF_SCOPE"] This approach ensures that your sensitive and environment-specific data is decoupled from the codebase, enhancing maintainability and security.   Step 4: Configure OAuth Flow in Streamlit Define OAuth2 Component and Redirect Logic With your environment variables and secrets in place, it’s time to configure the OAuth flow in Streamlit using the OAuth2Component from the streamlit-oauth package. This component handles user redirection to the Salesforce login page, token retrieval, and response parsing upon return to your app. from streamlit_oauth import OAuth2Component oauth2 = OAuth2Component( client_id=CLIENT_ID, client_secret=CLIENT_SECRET, authorize_url=AUTHORIZE_URL, token_url=TOKEN_URL, redirect_uri=REDIRECT_URI ) # create a button to start the OAuth2 flow result = oauth2.authorize_button( name="Log in", icon="https://www.salesforce.com/etc/designs/sfdc-www/en_us/favicon.ico", redirect_uri=REDIRECT_URI, scope=SCOPE, use_container_width=False ) This button initiates the OAuth2 flow and handles redirection transparently. Once the user logs in successfully, Salesforce redirects them back to the app with a valid token.   Handle Session State for Tokens and User Data After authentication, the returned tokens are stored in st.session_state to maintain a secure, per-user context. Here’s how to decode the token and extract user identity details: if result: #decode the id_token and get the user's email address id_token = result["token"]["id_token"] access_token = result["token"]["access_token"] # verify the signature is an optional step for security payload = id_token.split(".")[1] # add padding to the payload if needed payload += "=" * (-len(payload) % 4) payload = json.loads(base64.b64decode(payload)) email = payload["email"] username = payload["name"] #storing token and its parts in session state st.session_state["SF_token"] = result["token"] st.session_state["SF_user"] = username st.session_state["SF_auth"] = email st.session_state["SF_access_token"]=access_token st.session_state["SF_id_token"]=id_token st.rerun() else: st.write(f"Congrats **{st.session_state.SF_user}**, you are logged in now!") if st.button("Log out"): cleartoken() st.rerun() This mechanism ensures that the authenticated user context is preserved across interactions, and sensitive tokens remain protected within the session. The username displays in the UI after a successful login. 😀         Step 5: Create and Expose SnapLogic Triggered Task Build Backend Logic with AgentCreator Snaps With user authentication handled on the frontend, the next step is to build the backend business logic using SnapLogic AgentCreator. This toolkit lets you design AI-powered pipelines that integrate with data sources, perform intelligent processing, and return contextual responses. You can use pre-built Snaps (SnapLogic connectors) for Salesforce, OpenAI, and other services to assemble your Sales Agent pipeline.   Generate the Trigger URL for API Access Once your pipeline is tested and functional, expose it as an API using a Triggered Task: In SnapLogic Designer, open your Sales Agent pipeline. Click on “Create Task” and choose “Triggered Task”. Provide a meaningful name and set runtime parameters if needed. After saving, note the generated Trigger URL—this acts as your backend endpoint to which the Streamlit app will send requests. This URL is the bridge between your authenticated frontend and the secure AI logic on SnapLogic’s platform. However, before connecting it to Streamlit, you'll need to protect it using SnapLogic API Management, which we'll cover in the next section.   Step 6: Secure API with SnapLogic API Manager Introduction to API Policies: Authentication and Authorization To prevent unauthorized access to your backend, you must secure the Triggered Task endpoint using SnapLogic API Management. SnapLogic enables policy-based security, allowing you to enforce authentication and authorization using Salesforce-issued tokens. Two primary policies will be applied: Callout Authenticator and Authorize By Role. The new Policy Editor of SnapLogic APIM 3.0 Add Callout Authenticator Policy This policy validates the access token received from Salesforce. Since Salesforce tokens are opaque (not self-contained like JWTs), the Callout Authenticator policy sends the token to Salesforce’s introspection endpoint for validation. If the token is active, Salesforce returns the user's metadata (email, scope, client ID, etc.). Example of a valid token introspection response: { "active": true, "scope": "id refresh_token openid", "client_id": "3MVG9C...", "username": "mpentzek@snaplogic.com", "sub": "https://login.salesforce.com/id/...", "token_type": "access_token", "exp": 1743708730, "iat": 1743701530, "nbf": 1743701530 } If the token is invalid or expired, the response will simply show: { "active": false } Below you can see the configuration of the Callout Authenticator Policy: Extract the domain from the username (email) returned by the Introspection endpoint after successful token validation for use in the Authorize By Role Policy. Add AuthorizeByRole Policy Once the token is validated, the Authorize By Role policy inspects the username (email) returned by Salesforce. You can configure this policy to allow access only to users from a trusted domain (e.g., @snaplogic.com), ensuring that external users cannot exploit your API. For example, you might configure the policy to check for the presence of “snaplogic” in the domain portion of the email. This adds a second layer of security after token verification and supports internal-only access models.   Step 7: Connect the Streamlit Frontend to the Secured API Pass Access Tokens in HTTP Authorization Header Once the user has successfully logged in and the access token is stored in st.session_state, you can use this token to securely communicate with your SnapLogic Triggered Task endpoint. The access token must be included in the HTTP request’s Authorization header using the Bearer token scheme. headers = { 'Authorization': f'Bearer {st.session_state["SF_access_token"]}' } This ensures that the SnapLogic API Manager can validate the request and apply both authentication and authorization policies before executing the backend logic.   Display API Responses in the Streamlit UI To make the interaction seamless, you can capture the user’s input, send it to the secured API, and render the response directly in the Streamlit app. Here’s an example of how this interaction might look: import requests import streamlit as st prompt = st.text_input("Ask the Sales Agent something:") if st.button("Submit"): with st.spinner("Working..."): data = {"prompt": prompt} headers = { 'Authorization': f'Bearer {st.session_state["SF_access_token"]}' } response = requests.post( url="https://your-trigger-url-from-snaplogic", data=data, headers=headers, timeout=10, verify=False # Only disable in development ) if response.status_code == 200: st.success("Response received:") st.write(response.text) else: st.error(f"Error: {response.status_code}") This fully connects the frontend to the secured backend, enabling secure, real-time interactions with your generative AI agent.   Common Pitfalls and Troubleshooting Handling Expired or Invalid Tokens One of the most common issues in OAuth-secured applications is dealing with expired or invalid tokens. Since Salesforce access tokens have a limited lifespan, users who stay inactive for a period may find their sessions invalidated. To address this: Always check the token's validity before making API calls. Gracefully handle 401 Unauthorized responses by prompting the user to log in again. Implement a token refresh mechanism if your application supports long-lived sessions (requires refresh token configuration in Salesforce). By proactively managing token lifecycle, you prevent disruptions to user experience and secure API communications.   Debugging OAuth Redirection Errors OAuth redirection misconfigurations can block the authentication flow. Here are common issues and their solutions: Incorrect Callback URL: Ensure that the SF_REDIRECT_URI in your .env file matches exactly what’s defined in the Salesforce Connected App settings. Missing Scopes: If the token does not contain expected identity fields (like email), verify that all required scopes (openid, email, profile) are included in both the app config and OAuth request. Domain Restrictions: If access is denied even after successful login, confirm that the user’s email domain matches the policy set in the SnapLogic API Manager. Logging the returned error messages and using browser developer tools can help you pinpoint the issue during redirection and callback stages.   Best Practices for Secure AI Application Deployment Rotate Secrets Regularly To reduce the risk of secret leakage and potential exploitation, it's essential to rotate sensitive credentials—such as CLIENT_ID and CLIENT_SECRET—on a regular basis. Even though Streamlit’s Secrets Management securely stores these values, periodic rotation ensures resilience against accidental exposure, insider threats, or repository misconfigurations. To streamline this, set calendar reminders or use automated DevSecOps pipelines that replace secrets and update environment files or secret stores accordingly.   Monitor API Logs and Auth Failures Security doesn’t stop at implementation. Ongoing monitoring is critical for identifying potential misuse or intrusion attempts. SnapLogic’s API Management interface provides detailed metrics that can help you: Track API usage per user or IP address. Identify repeated authorization failures or token inspection errors. Spot anomalous patterns such as unexpected call volumes or malformed requests.   Extending the Architecture Supporting Other OAuth Providers (Google, Okta, Entra ID) While this tutorial focuses on Salesforce as the OAuth 2.0 Identity Provider, the same security architecture can be extended to support other popular providers like Google, Okta, and Microsoft Entra ID (formerly Azure AD). These providers are fully OAuth-compliant and typically offer similar endpoints for authorization, token exchange, and user introspection. To switch providers, update the following in your .env file: SF_AUTHORIZE_URL SF_TOKEN_URL SF_SCOPE (as per provider documentation) Also, make sure your app is registered in the respective provider’s developer portal and configured with the correct redirect URI and scopes.   Adding Role-Based Access Controls For larger deployments, simple domain-based filtering may not be sufficient. You can extend authorization logic by incorporating role-based access controls (RBAC). This can be achieved by: Including custom roles in the OAuth token payload (e.g., via custom claims). Parsing these roles in SnapLogic’s AuthorizeByRole policy. Restricting access to specific APIs or features based on user roles (e.g., admin, analyst, viewer). RBAC allows you to build multi-tiered applications with differentiated permissions while maintaining strong security governance.   Conclusion Final Thoughts on Secure AI App Deployment Securing your generative AI applications is no longer optional—especially when they’re built for enterprise use cases involving sensitive data, customer interactions, and decision automation. This tutorial demonstrated a complete security pattern using SnapLogic AgentCreator and Streamlit, authenticated via Salesforce OAuth 2.0 and protected through SnapLogic API Management. By following this step-by-step approach, you ensure only verified users can access your app, and backend APIs are shielded by layered authentication and role-based authorization policies. The same architecture can easily be extended to other providers or scaled across multiple AI workflows within your organization.   Resources for Further Learning SnapLogic Resources and Use Cases Salesforce Developer Docs Streamlit Documentation OAuth 2.0 Official Specification With a secure foundation in place, you’re now empowered to build and scale powerful, enterprise-grade AI applications confidently.