AgentForge4j
The embeddable Java framework for governed AI workflows.
AgentForge4j is an embeddable, open-source Java framework for designing and running human-in-the-loop AI workflows. Workflow structure, transitions, agents, prompts, and approval points can be defined externally — the runtime follows the orchestration you authored rather than inventing it at runtime.
Who it's for
AgentForge4j is for Java teams building their own AI-powered business or engineering workflows — not simply using an off-the-shelf AI assistant. It is useful when AI must follow controlled steps, call internal tools, pause for human approval, retry safely, and leave an auditable record of what happened.
Typical use cases include document review, compliance and onboarding processes, support triage, and internal engineering or security checks.
Predictable, not improvised
The workflow author decides which agent runs at each step at design time. The runtime follows that authored orchestration rather than choosing agents dynamically, and every run is observable through the events and state it emits — though the LLM’s own output can still vary between runs.
Externally defined, not hardcoded
Workflow structure, transitions, agents, and system prompts can be defined in external JSON and Markdown files, so a workflow’s orchestration can change without changing code. Applications may still provide their own code-defined tools, integrations, providers, persistence, and extension implementations.
Human approval where it matters
Transition gates (automatic, human-review, or human-approval) sit between steps, so a workflow can pause for a real decision exactly where the author placed one — not everywhere, and not nowhere.
Framework 0.1.0 and Workflow Catalog 0.1.0 are publicly available. The project remains pre-1.0 and under active development; see the catalogue for what ships today.