Skip to main content
The lab produces a focused set of projects that work together as an agent engineering stack β€” for building, evaluating, and understanding AI agents on the JVM.

Loopy

Loopy

A small agentic coding CLI for Java developers β€” an agent loop, a terminal UI, and multi-provider support, with nothing else bolted on.

AgentWorks

Core agent-* libraries that compose into a complete agent engineering stack. Use the AgentWorks BOM for coordinated version management.

Agent Workflow

Compose multi-step agentic pipelines β€” steps, typed context, branching, loops, quality gates, parallel execution. Compiles to a graph IR with portable runtimes.

Agent Client

Autonomous CLI agent integrations β€” a unified framework actively verified with Claude Code, Codex, and Gemini CLI. Other adapters remain experimental. Portable MCP servers, auto-configuration, and reasoning effort (low/medium/high) across providers, with provider-native ranges when finer control is needed.

Agent Judge

Agent-agnostic evaluation with deterministic, command, and LLM judges. Jury system with voting strategies. Zero cost for T0-T1 checks β€” LLM fires only when cheaper tiers pass.

Agent Bench

Benchmarking suite for Java-centric AI agents on enterprise dev tasks β€” issue triage, PR review, coverage, compliance. Agent-agnostic with provide/grade separation.

Agent Experiment

End-to-end experiment driver β€” datasets, runner, jury scoring, variant comparison, sessions, and sweeps. The execution backbone of every lab experiment.

Agent Memory

Progressive memory management β€” context compaction, token-budgeted retrieval, LLM summarization. Ships as a Spring AI BaseAdvisor; Tier-1 compaction today, tiers 2-4 planned.

Agent Hooks

Portable hook API for steering agent behavior at the tool-call boundary β€” deterministic safety, observability, and input modification that works across any runtime.

Agent Journal

Behavioral trace capture β€” every tool call, state transition, and decision point. Feeds Markov fingerprinting and cross-variant behavioral comparison.

Agent Sandbox

Isolated command execution β€” local, Docker, or E2B cloud backends behind a unified API. Safe execution for untrusted agent code.

Agent Tools

Claude Code-inspired tools for Spring AI agents β€” file I/O, shell, search, web, subagent orchestration.

Agent Skills

Curated domain knowledge modules (SkillsJars) β€” teach agents how to test, review, and build Spring applications.

SDKs & Protocols

Foundational SDKs for agent communication and CLI integration.

Claude Agent SDK (Java)

Java SDK for Claude Code CLI β€” three-API architecture, 23-part tutorial, Spring Boot auto-configuration

ACP Java SDK

Agent Communication Protocol β€” build agents, consume agents, integrate with Zed, JetBrains, and VS Code