Joget

Agents have a Database Query Tool, JSON API Tool, Email Tool, Google Web Search Tool, BeanShell Tool for custom code, Run AI Agent Tool for calling other agents, MCP Client Tool for remote MCP servers and Pinecone Tool, covering the customer's database, any API, messaging, the open web, custom code, agent-to-agent calls, MCP services and a vector store.

Underneath, the platform's Application and Integration Fabric connects legacy and modern systems through prebuilt connectors and an extensible plugin architecture with an open-source core. No connector catalog or count is published for the platform integration layer.

Sourcekb.joget.org AI Agent Builder Properties documentation, joget.com product pagesread 2026-08-31

An agent has a brain that sets the LLM service and default prompts, plus Tasks that run in sequence, each with a Prompt, an optional Tool and an optional Enhancer. Decisions, a Response Condition and a BeanShell Decision, branch on the agent's output, Async Agent Execution runs agents in the background, and a Run AI Agent Tool lets one agent call another. The AI Agent Tool embeds agents as steps in Joget's process engine, so agent tasks, human tasks and deterministic routing combine in one process, with process variables carrying state.

Sourcekb.joget.org AI Agent Builder navigation and AI Agent Tool documentationread 2026-09-29

Agents can use a Pinecone Tool with first-class integration to a persistent vector index, alongside a Database Query Tool and Google Web Search Tool, and prompts include File Prompt for documents, Form Data Prompt and List Data Prompt for application records, Form Data Tables Schema Prompt for the data model, and JSON API Prompt for live external data.

Grounding covers documents, the application's own structured records and schema, and external sources. Chunking, the embedding model, index maintenance and answer citation are not described, and Pinecone is a dependency the customer supplies, not a bundled store.

Sourcekb.joget.org AI Agent Builder Properties documentationread 2026-08-31

The AI Agent Tool embeds agents as steps in Joget's workflow and process engine, whose human activity step assigns work to a person, so agent tasks and human approval tasks sit as steps in one process. Inside the agent, Guards include a Rule-Based Guard, an LLM Safety Check and a Webhook Guard that calls an external service before the agent proceeds, and Decisions route on the response. DX 8's security audit inspects process participation. The approval step is the process engine's human task, not a pause inside the agent.

Sourcekb.joget.org AI Agent Builder navigation and AI Agent Tool documentationread 2026-09-29

Joget is ISO/IEC 27001:2022 certified. Customer controls include role-based access control and audit trails as core platform features, a Governance Center for managing and monitoring installations, apps and developers, built-in monitoring in DX 9, and one-click security, performance and governance audits from DX 8 covering user authentication, user interface access and process participation, plus checks for unusually high error rates and data integrity in running processes. FISMA, PCI DSS and DISA compliance claims appear only on a partner integrator's site.

Sourcejoget.org DX, DX 9 and DX 8 pages, kb.joget.org Governance Center documentationread 2026-08-31

An Agent Execution Audit Trail, available from AI Agent Builder 8.2.0, tracks agent activities, and System Agents and Traceability sits alongside Async Agent Execution. The platform has built in monitoring, audit trails and a Governance Center, and agents run as steps inside governed processes, so their actions sit in the same process history as human tasks. The fields each execution record holds are not spelled out.

Sourcekb.joget.org AI Agent Builder navigation and audit trail pageread 2026-09-29

A Mem0 Chat Memory Tool lets an agent store and recall context, user preferences and relevant information for each use, and Store to Form and Store to Process Variable enhancers write agent output into application data and process variables for later steps. Agents can write memory, but no scope, lifetime, purge path or way to review or delete memories is described. Application tables and process variables are storage that the agent's work reads.

Sourcekb.joget.org AI Agent Builder navigation and Mem0 Chat Memory Tool pageread 2026-09-29

Joget deploys anywhere, containerized, cloud-native or on-premise, and runs across private cloud, public cloud, hybrid and on-premises, with containerization and certified cloud deployments on AWS, Azure, Red Hat and Google Cloud. An open-source Community Edition allows self-hosting without a vendor engagement.

Agents use a customer-supplied LLM key with an optional proxy domain that routes model calls through the customer's own gateway, so an on-premises deployment need not send prompts through a vendor-managed inference path. The AI Agent Builder is an Enterprise Edition feature from DX 8.1 and is included by default from DX 9.1. Region selection within Joget's own cloud offering is not described.

Sourcejoget.org home and DX pages, kb.joget.org AI Agent Builder documentationread 2026-08-31

The Joget Marketplace is an app store for enterprise apps, plugins and templates, and the AI Agent Builder plugin itself is distributed and installed through it. Within the builder, reusable components such as a Persona Prompt for defining an agent's role, plus a best practices page for prompting and designing agents, speed up construction. There is no library of ready made agents, agent template gallery or packaged departmental agents, since the marketplace holds apps, plugins and templates, not prebuilt agents.

Sourcekb.joget.org AI Agent Builder and Properties documentation, joget.org marketplace and DX pagesread 2026-08-31

The AI Agent Tool places an agent as a step in a Joget business process, so agents start with whatever starts the process, and the process engine supports scheduled, event driven, form submission and API triggered starts. An Email Tool provides an outbound channel, and agents embedded in Joget applications get an inbound surface through those applications' interfaces. No scheduling or cron configuration, inbound webhook, event subscription, or messaging or chat channel is listed for AI agents specifically, and the agent material does not restate the platform's trigger types.

Sourcekb.joget.org AI Agent Tool and AI Agent Builder documentation, joget.com AI Agent Builder pageread 2026-08-31

The AI Agent Builder's LLM Services are OpenAI, Anthropic, Google Gemini, OpenRouter LLM and IBM. Installing it needs an LLM key for Anthropic Claude, OpenAI or custom models, and the model is set per agent in the Define Agent section under a bring-your-own-key arrangement. A proxy domain setting routes model calls through the customer's own gateway.

OpenRouter fronts many providers and custom models are supported, so the set of reachable models is not limited to Joget's own integrations. Older marketplace knowledge base pages that say OpenAI is the only supported model describe a superseded build and are still live.

Sourcekb.joget.org AI Agent Builder and AI Agent Builder Properties documentationread 2026-08-31

The platform offers APIs for integration and a plugin architecture with an open-source core, and the AI Agent Builder itself ships as a plugin JAR downloaded from the Joget Marketplace and installed through Manage Plugins. BeanShell Prompt, Tool and Enhancer components let custom code run as a first-class part of an agent.

White label and OEM distribution are offered. The MCP Client Tool calls remote MCP servers, but Joget publishes no MCP server of its own that exposes agents to outside assistants, and no API reference for invoking an agent from outside, as distinct from the platform's general APIs.

Sourcekb.joget.org AI Agent Builder and Properties documentation, joget.org DX and home pagesread 2026-08-31

Builders test an agent in the Preview tab after configuring tasks and before saving it into a workflow, simulating scenarios, watching the task sequence and refining prompts, tools and enhancers before the agent reaches a live process. The prompt, tool and enhancer structure shows which step produced a given output. No scoring, judge, retained test set, recorded expected outputs, regression harness or comparison between agent versions is described.

Sourcekb.joget.org AI Agent Builder documentation, joget.com AI Agent Builder pageread 2026-08-31

The agent tools are BeanShell, Database Query, Email, Google Web Search, JSON API, Run AI Agent, Mem0 Chat Memory, MCP Client and Pinecone, all programmatic interfaces, and none drives a browser or an interface. The Google Web Search Tool returns results over an API, which is retrieval, not operation. A customer could extend the BeanShell Tool toward browser automation with custom code, but that is not a Joget capability. No browser control, navigation, form filling, screen interaction, visual grounding or desktop automation is described, and Joget makes no claim in that area.

Sourcekb.joget.org AI Agent Builder Properties documentationread 2026-08-31