SoftTech AI Registry
SnapLogic · SnapLogic AgentCreator

Agent tool pipeline execution

AgentCreator agents reason over the tools made available to them and, when the model determines a tool is needed, invoke SnapLogic tool pipelines that can retrieve data, calculate, or perform persistent operations such as database writes and sending email, including against systems outside SnapLogic. Tool pipelines run using configured SnapLogic Accounts (connection credentials).

Recorded characteristics

Function
SnapLogic documents the Agent Snap as the core of an Agent pipeline that manages the full cycle of receiving a request, reasoning through it, calling tools and returning a response. When the LLM determines that a tool is needed, the Agent Snap invokes the appropriate tool pipeline, passes the required arguments and feeds the result back into an iterative reasoning loop, which continues until the objective is met. SnapLogic states this is the mechanism that gives agents the ability to act on external systems rather than simply generate text. Function Generator Snaps produce tool definitions describing which tools are available and what parameters they require, so the LLM can reason about which tool to invoke; a Multi Pipeline Function Generator can expose multiple tool pipelines. A pipeline becomes a tool by enabling "Expose this pipeline as a tool" and entering a plain-language tool description and tool inputs. Agent pipelines may call other Agent pipelines as subagents. AgentCreator agents may also call external MCP servers through the MCP Client Snaps (MCP Function Generator, MCP Invoke) as part of their tool surface.
Data access
Tool pipelines can be built around any Snap that performs the required operation, including an HTTP Client Snap for external web services and system-specific Snaps such as a MySQL Snap for database operations. SnapLogic describes each tool pipeline as a discrete capability such as retrieving data, querying a system, performing a calculation or writing a result. Data reached by an agent is therefore determined by the tool pipelines and the SnapLogic Accounts they are configured with. Each tool pipeline must return exactly one output document per invocation.
Actions
Can take actions
External actions
Yes
Human confirmation
Conditional
Permission basis
Separate permissions
Administrative control
Documented controls are builder-level: which pipelines are exposed as tools; tool descriptions and input schemas; which Accounts each Snap uses (Accounts are accessible to their creator and to pipelines in the Project folder where created, or to anyone with access to a shared folder); administrator-created service accounts; dynamic accounts supplied by pipeline parameters or a Secrets Manager; an adjustable per-tool-pipeline timeout in the Agent Snap; and inspection of reasoning steps and tool call history in the Agent Visualizer. Pipeline validation, testing and Agent Visualizer inspection are not confirmation mechanisms. No AgentCreator-specific administrator enable/disable switch was established.
Default state
Conditional
Availability
AgentCreator is a current SnapLogic feature within SnapLogic Designer, presented without preview or beta labelling in the reproduced documentation. An agent can take actions only after it is built: an AgentCreator feature subscription, construction of an Agent pipeline and tool pipelines, exposing pipelines as tools with descriptions, configuration of connection Accounts, and deployment as a Triggered Task or Ultra Task for production use. Access to AgentCreator does not by itself make any agent operational. No exact general-availability date is established.
Licensing
SnapLogic lists a feature subscription to AgentCreator as a prerequisite for building tool pipelines. No further entitlement detail is recorded.
External model or provider
Customer-selected; no single model or provider. SnapLogic documents vendor-specific Agent Snaps (Amazon Bedrock Converse API Agent, Azure AI Foundry Chat Completions Agent, Google Gemini API Agent, OpenAI Chat Completions Agent) and Function Generator Snaps for Amazon Bedrock, Google Gemini, OpenAI and Azure OpenAI. Functionality is not identical across every vendor; see the SnapLogic support matrix.
Limitations and uncertainty
(1) Permission basis is recorded as separate_permissions: tool pipelines execute using separately configured SnapLogic Accounts/connection credentials, not by established passthrough of the interacting user's permissions. No reproduced evidence establishes that any connector passes through the interacting user's own identity to the target system; the agent is not established as having a dedicated identity of its own. (2) Human confirmation is conditional: SnapLogic documents Human-in-the-Loop checkpoints that pause execution until a human supervisor authorizes a high-stakes action, and its September 2026 release references human-in-the-loop steps within tool pipelines, but no universal built-in approval gate before every tool invocation is documented; the Agent Snap invokes tools when the model determines they are needed, so approval depends on agent and pipeline design. (3) Not every tool is state-changing; retrieval, query, calculation and generated output are distinct from persistent action. External persistent action rests on SnapLogic's documentation of tool pipelines writing to databases, calling web services via HTTP Client and the first-party "AgentCreator Tool - Send email" and "PostgreSQL Execution Agent" tool patterns; the pattern pages require login and their internal steps were not reproduced. (4) Boundary: excludes external AI agents or clients invoking SnapLogic pipelines through the SnapLogic MCP Server, which SnapLogic distinguishes as a separate direction (future candidate), and excludes SnapGPT, a design-time helper that can generate agent and MCP Server pipelines. (5) Unresolved: exact GA date; any AgentCreator-specific administrator enablement switch.

Evidence