SnapLogic

The platform provides more than a thousand prebuilt connectors called Snaps spanning cloud applications, on-premises business software, databases, data warehouses, ERP, CRM and SaaS products, all reachable by agents because agents are built as pipelines from the same Snaps.

Tool calling is documented through Tool Calling Snaps across the LLM Snap Packs, OpenAI Responses API tool calling, and an OpenAPI Function Generator Snap that converts an OpenAPI specification into callable functions. The 2026 platform added tool lifecycle management, versioning, metadata and lineage controls governing the tools agents invoke.

Sourcedocs.snaplogic.com AgentCreator documentation and August 2025 release notes, snaplogic.com AgentCreator product pageread 2026-08-31

The vendor documents pipelines as the orchestration primitive, with customers orchestrating a dynamic sequence of pipeline runs to accomplish a complex task and deploying it to production as a Task, and AgentCreator combining dynamic iteration with real-time generative decision-making so the model determines the path while the pipeline engine executes it.

Multi-agent coordination is documented as workflows coordinating activity across multiple agents and enterprise systems simultaneously. Control flow is inherited from the integration engine, including sub-pipeline invocation through the Pipeline Execute Snap, error views on all Snaps, configurable timeouts, and resumable pipelines with a suspended state carrying notification settings. Ultra Tasks support always-on pipelines running continuously.

Sourcedocs.snaplogic.com AgentCreator, task and ultra pipeline documentation, snaplogic.com AgentCreator product pageread 2026-08-31

AgentCreator ships vector database Snap Packs supporting Pinecone, OpenSearch, MongoDB and Snowflake, with utilities for parsing HTML, Markdown, PDF and unstructured data and for embedding and storing that data in vector databases, so a customer builds and maintains a persistent retrieval index.

An Agent Retrieve and Generate Snap makes retrieval-augmented generation a named platform operation, and OpenAI and Azure OpenAI vector store management with file uploading is documented as improving contextual awareness of future queries. Grounding reaches operational systems as well as documents through the platform's prebuilt connectors to ERP, CRM, databases and warehouses, and the IDP Agent provides prebuilt patterns for invoice and purchase order processing. Chunking, re-indexing and index freshness are managed by the customer since the vector store is theirs.

Sourcedocs.snaplogic.com AgentCreator documentation and August 2025 release notes, snaplogic.com AgentCreator product pageread 2026-08-31

AgentCreator documents enterprise-grade governance comprising AI response validation and fine-grained access control, which constrain agent output and reach. No runtime approval step, checkpoint, pause-for-confirmation, pending-action queue or human reviewer surface is documented.

The July 2026 Handle Errors via Agent toggle passes downstream tool failures back to the model for self-correction, which is autonomous recovery rather than human oversight, and the vendor's product page advertises resolving inquiries without human intervention.

Sourcedocs.snaplogic.com AgentCreator documentation, snaplogic.com AgentCreator product page and data sheet, snaplogic.com July 2026 release blogread 2026-08-31

The vendor's security and compliance page states the SnapLogic Platform for Agentic Integration is certified by third parties with SOC 1 Type 2 under ISAE 3402 and SSAE 22, SOC 2 Type 2, SOC 3 and HIPAA-HITECH, and is compliant with CCPA and GDPR, with the security standards page confirming annually renewed third-party audits and audit summary letters available on request under NDA.

Named customer-facing controls include single sign-on through SAML, integration with IAM and identity provider systems including Okta, Ping and OpenAM, role-based access controls, per-asset permission tiers, encryption in transit and at rest, and enhanced account encryption and secrets management. AgentCreator adds fine-grained access control and AI response validation. The vendor states it does not store customer data or pipeline content, only metadata held within the control plane. The ISO 27001, 27017 and 27018 certifications named on the same page belong to AWS as the underlying infrastructure provider rather than to the vendor.

Sourcesnaplogic.com security-compliance and security-standards pages, docs.snaplogic.com task permissions documentationread 2026-08-31

AgentCreator ships an Agent Visualizer that the vendor describes as giving transparent, auditable insight into agent reasoning and decisions, with builders able to inspect reasoning steps, tool calls and results.

Because agents are built as pipelines, a pipeline can hold the entire conversational history from the system prompt through all user and assistant messages, documented as enabling debugging of the individual operations performed at each runtime iteration.

The platform adds tool lifecycle management, versioning, metadata and lineage controls, and the SnapLogic Dashboard provides pipeline and task execution monitoring with per-instance run details and failure states.

The Agent Visualizer is described in design-time terms, and production reasoning-trace retention, retention periods and export to an external observability stack are not documented.

Sourcesnaplogic.com AgentCreator data sheet and product page, docs.snaplogic.com AgentCreator and task documentationread 2026-08-31

AgentCreator documents that a pipeline can hold the entire conversational history from the system prompt through all user and assistant messages, which is state within a run. For durable conversation state the vendor documents delegating to the model provider, writing that another system manages it automatically, with OpenAI and Azure OpenAI support extending to management of vector stores and file uploading to improve contextual awareness of future queries; that persistence is the provider's mechanism reached through a Snap rather than a platform memory object.

No SnapLogic-native memory Snap, agent memory store, session identity object, cross-session context primitive or retention configuration is documented. The vector database Snap Packs serve retrieval grounding rather than agent memory.

Sourcedocs.snaplogic.com AgentCreator documentation, snaplogic.com AgentCreator product page and data sheetread 2026-08-31

The platform separates a vendor-hosted control plane from a customer-controlled execution plane called a Snaplex, which runs as a Groundplex inside the customer's own network and firewall, as a Cloudplex, or as a hybrid mix. The documentation shows a Triggered Task routed through the customer's on-premises load balancer to a Groundplex node, which contacts the control plane only to retrieve assets and can execute from cache thereafter.

The vendor states it does not store customer data or pipeline content in the platform, only metadata held within the control plane with no outside service access, so data location follows the customer's Snaplex placement. Agents built in AgentCreator are pipelines and execute on the same Snaplex. A separate EMEA environment is documented at platform level.

Sourcesnaplogic.com security-compliance page, docs.snaplogic.com triggered-tasks and scheduled-tasks documentationread 2026-08-31

The SnapLogic Public Pattern Library contains AgentCreator pipeline patterns that customers filter for by selecting AgentCreator and GenAI Builder as search criteria, with patterns documented as canonical pipeline designs that can be repurposed in the customer's own environment.

The GenAI package ships an IDP Agent automating document processing with prebuilt patterns for named tasks including invoice and purchase order handling, alongside bundled utilities for parsing and embedding, and Prompt Composer provides prompt development with SnapGPT assistance. No marketplace of finished role-based agents is documented; the patterns are designs to adopt rather than agents to activate. The prebuilt Snaps are connectors rather than agents.

Sourcedocs.snaplogic.com AgentCreator documentation, snaplogic.com AgentCreator product pageread 2026-08-31

The vendor's task documentation names three task types that make pipelines and agents operational: Scheduled Tasks running at a specific time, on an interval, or on a cron schedule with calendar dates, clock time, frequency, end dates, time zones and blackout dates for planned outages; Triggered Tasks exposing a pipeline as a web API endpoint invoked over HTTP with data passed in and returned, supported by the Open API specification and downloadable in API form; and Ultra Tasks for low-latency always-on jobs, where a FeedMaster node queues incoming messages and a listener-consumer pipeline continuously polls a messaging service such as a JMS consumer. Triggered Tasks can be routed through an on-premises load balancer to a Groundplex node so the endpoint sits inside the customer's network.

Sourcedocs.snaplogic.com task, scheduled-tasks, triggered-tasks and ultra task documentationread 2026-08-31

AgentCreator ships LLM Snap Packs connecting to Claude, OpenAI, Azure OpenAI, Google Gemini and Amazon Bedrock, each configured as a separate account the builder selects, alongside vector database Snap Packs for Pinecone, OpenSearch, MongoDB and Snowflake. The vendor's security and compliance page states the platform is agnostic regarding AI and LLM models, giving customers flexibility to choose solutions meeting their business and compliance needs.

OpenAI custom endpoint accounts, in plain, OAuth2 and authorization-header variants, route model calls through an API management proxy so an organization can control model access by policy, and an AI gateway added in April 2026 provides consistent support across providers. Model selection is per-Snap configuration rather than runtime routing between models.

Sourcedocs.snaplogic.com AgentCreator documentation and August 2025 release notes, snaplogic.com AgentCreator product page and security-compliance pageread 2026-08-31

The vendor documents an MCP Server feature that exposes SnapLogic pipelines as AI tools any MCP-compatible client can discover and invoke, stating explicitly that it makes pipelines available to external AI agents and clients such as Claude, custom AI applications and API testing tools, as complementary to AgentCreator's internally-running agents.

Triggered Tasks expose any pipeline as an HTTP web API endpoint with Open API specification support and endpoints downloadable in API form, SnapLogic Public APIs are documented for platform operations, API management is a first-class product capability, and custom Snaps let developers build and package their own components. SnapCode was added in 2026. Triggered Tasks can be served from a Groundplex behind the customer's own load balancer.

Sourcedocs.snaplogic.com AgentCreator, triggered-tasks and run-triggered-task documentation, snaplogic.com July 2026 release blogread 2026-08-31

The vendor's AgentCreator Evaluation tutorial documents an evaluation pipeline pattern that reads a control set of prompts with expected answers, runs each through the customer's retrieval or agent pipeline via the Pipeline Execute Snap, scores every response with reasoning, and writes the scores back, with guidance to compare subtle differences between LLM pipelines, find the categories where the application is accurate, and rerun periodically to catch model drift.

Two evaluation patterns, for Azure OpenAI and Amazon Bedrock, ship in the Public Pattern Library. Debugging adds the Agent Visualizer with reasoning steps and tool calls and Prompt Composer.

Sourcedocs.snaplogic.com AgentCreator Evaluation tutorial and AgentCreator documentationread 2026-09-29

Agents are built as pipelines from Snaps, which connect to third-party API endpoints, so all agent action runs through documented programmatic interfaces. No browser control, navigation, form filling, screen interaction, visual grounding or desktop automation is documented, and the vendor makes no claim in that territory. Utilities that parse HTML alongside PDF, Markdown and unstructured content perform document parsing rather than browser operation.

Sourcedocs.snaplogic.com AgentCreator and Snap Pack documentation, snaplogic.com AgentCreator product page and data sheetread 2026-08-31