Clear tools and reasoning
Agent can only call app actions you explicitly connected - no hidden tool list, no surprise API calls.
Build AI Agent workflows you can control
WEXTL® AI Agent node [1] lets you build autonomous agents on the same visual canvas with your apps. Integrate 1,300+ apps [2] with AI - use them as tools agent can call. Platform supports Memory and Knowledge hubs, and outputs of AI Agent node can be gated with regular Condition, Merge or Filter nodes [3].
WEXTL® turns AI agents into transparent, accountable systems — every tool call and decision the agent makes is logged and visible on the run history [4], easy to audit and easy to manage.
- GDPR
- Data Encryption
- 2FA
- Local Data Region
Video slide shows how AI Agent node is used on WEXTL® canvas.
Memory scoped per entity
Templated memory keys reuse context per customer or per ticket automatically without cross-tenant leaks.
Access to knowledge
Knowledge hub gives to AI agent node a search tool over your docs - no separate retrieval pipeline to build.
Guardrails for responses
Filter node on agent output - move further safe cases only. The rest can be given to Human in the Loop node [6].
How to build AI Agent on WEXTL®?
There are only 3 main steps to create an AI agent [1]. Under 10 minutes [4] setup - from picking a trigger to a live agentic automation that operates per your guardrails, calls correct tools, and keeps context of its previous runs.
STEP 1
Select the trigger for AI agent
Choose how the workflow starts - with webhook [3], app event, or scheduled [5] trigger. The trigger runs and resolves data before the AI agent node [1] step receives an input, so AI always operate on real, validated inputs.
1300+ apps
STEP 2
Connect model and tools
Select the AI model to use and wire credential, then connect apps to the Tools hub - so AI agent can call them [4]. Use Memory or Knowledge hubs to give it context during execution.
STEP 3
Route based on output
Add a Condition or Merge node after the agent - route the safe path straight to eg. Slack or Gmail, and everything uncertain waits for a human review. Same visible decision logic [3] as the rest of the builder.
How does the AI Agent choose which tools to call?
Every tool the AI Agent [1] can call has to be explicitly attached to it via tools hub. There is no hidden tools list [4] - the AI model uses only what workflow builder instructed - so automator always in control.
WEXTL® AI agent node loops until done - it automatically calls necessary tools, sends outputs to model for evaluation and response. Each step agent does can be visually traced on the workflow canvas [7].
Knowledge and Memory [1] are used to give AI the context of the conversation. Memory key is the way to keep conversations with exact customer IDs - to prevent any context leaking from one customer flow to another. Even though Memory and Knowledge are not required to be attached to AI Agent node, it is a common practice [4] to use them.
Video shows an AI Agent answering a Slack question by calling only its attached tools, Pipedrive, Notion and Slack, on WEXTL®.
What is the best practice for using Agent memory and context?
Select an AI Memory tool to set a persistent or per-run scope, retention strategy, and rolling window size - same param panel as any other node [3] has. Memory key can be set from incoming node, eg. {{trigger.user_id}} - this will allow AI agent bucket messages per user ID [4]. Blank memory key still stores a conversation, but mixes the context of it with every new run.
Context is managed by Knowledge tool that can use your text data to answer user questions - no guessing, safely relying on provided knowledge. AI agent calls knowledge itself, automatically - no explicit prompt needed. When building automation, you just need to connect Knowledge to AI Agent and provide information if you want AI rely on for providing responses.
Video shows AI Agent Memory keyed by user ID, keeping each customer's history separate, and Knowledge answering from your own text on WEXTL®.
Supported integrations
Frequently asked questions
This page documents the AI Agent node's own configuration surface — its Model, Tools, Memory, and Knowledge hubs. The Solutions page shows deployable role workflows built on top of it, like inbox triage and meeting-brief agents.
Yes. Wire an AI Agent node into another agent's Tools hub as a sub-agent. Call depth is capped at five and a cycle guard blocks an agent from indirectly calling back into itself, so a mis-wired canvas fails one tool call instead of looping the run forever.
Yes. Wire a Database node into the Tools hub the same way as an app action. The agent decides when to query it, and each read or write shows up as its own logged step in run history.
Attach a Knowledge hub backed by an embeddings collection. WEXTL synthesizes a `knowledge_search` tool automatically — no separate retrieval pipeline to build. Multiple Knowledge hubs each get their own named tool over their own collection.
Clone an AI agents solution or template, wire an AI Model node and one or two Tools hub entries to start, and add Memory or Knowledge hubs only once the base loop is working end to end.
Put an agent on your canvas or explore more
References
List of resources used for writing this documentation page on WEXTL®. Follow a citation to jump straight to the source.
- WEXTL® AI Agent — node documentation.
- WEXTL® Apps Library — full catalog of supported app integrations.
- WEXTL® built-in workflow nodes — catalog of built-in nodes with per-node documentation pages.
- WEXTL® anonymized usage experience from customers — collected by support team and/or partners.
- WEXTL® Workflow Scheduler
- WEXTL® Human in the Loop node — pause a run for human review or approval before proceeding.
- WEXTL® visual workflow builder — canvas for building and visually tracing automations.