Describe a single Agent's model, instructions, tools, memory, compact, sandbox, and workflow parameters with declarative YAML. Support loading, validation, and templating, then create a runnable Agent once the host supplies its dependencies.
Source-map navigation for the ai4j-agent core classes: layered by wiring, runtime, model adaptation, tools, memory, workflow, subagent, team, and trace — tells you which line to read first and which entry point fits your question.
Navigate the Agent Teams runtime from the source view: how AgentTeamBuilder fields map to the execution chain, the task board state machine, the team tool surface, persistence recovery, and extension points.
Technical roadmap for ai4j-agent: phased evolution from the P0 runtime kernel (Session/Memory/Compact/Plugin/Permission) through P1 Blueprint YAML, P2 Sandbox SPI, P3 Coding sandbox routing, P4 CLI, and P5 Remote Runner.
Versioned Javadoc entry points for each AI4J Maven module, indexed by capability for Core SDK, Agent, Coding Agent, and extension SPI Java types and signatures.
Summary of the high-frequency slash commands implemented by the Coding Agent (provider/model/mcp/session/process/team/compact, etc.), including their scope, common parameters, and command-visibility differences across the CLI/TUI/ACP hosts.
Explains the three layers of Coding Agent configuration (global provider assets, global MCP definitions, workspace bindings), per-field resolution precedence, and which changes trigger a rebind of the current session runtime.
The canonical reading map of the AI4J docs: what question each of the eight top-level sections answers, the single entry point for every capability, and how old links redirect automatically.
Explains the two-layer internal SPI behind plugin discovery and resource reading: the ExtensionLoader interface lets you discover plugins without ServiceLoader (the default ServiceLoaderExtensionLoader is annotated @Internal), the ExtensionResourceResolver public helper reads classpath text resources in plugin classloader -> TCCL -> resolver classloader order and isolates each plugin jar, and the DiscoveredExtension inspection projection (manifest + extension + enabled) returned by ExtensionRegistry.list() after discovery.
A consolidated FAQ for the AI4J docs site: where to start, how the SDK relates to the AI foundation, and how to disambiguate concepts such as Chat/Responses, Function Call/Tool, MCP/Agent, Skill/Tool, and ACP/MCP.
AI4J feature map: lists each capability's status, owning module, suitable scenarios, and further-reading entry point by maturity marker (stable/advanced/preview/experimental), helping you adopt the minimal module on demand.
The task control-plane contract that external callers actually face: the five HTTP entry points validate/run/report/result/cancel, how DTOs are reshaped by the adapter and facade, and the configuration items that genuinely change external behavior.
Unified table of core AI4J terms: defines Agent, AiService, Chat, Coding Agent, Function Call, MCP, Memory, Skill, Responses, Tool Registry, and other key concepts to avoid cross-topic confusion.
Maps AI4J's historical CSDN blog articles to structured docs, so they can be maintained continuously and kept in sync across versions. Covers historical content such as Spring Boot + OpenAI quickstart wiring and DeepSeek/Qwen/Llama local model integration, along with the corresponding new doc paths.
A thorough walkthrough of ai4j's RAG answer-quality judge layer: the RagJudge SPI, the built-in ChatRagJudge three-dimension scoring protocol (faithfulness/contextRelevance/answerRelevance, temperature 0, json_object, fault-tolerant parsing, score clamping to [0,1]), how RagOnlineEvaluator writes the judgment into RagTrace, and where it draws the line against the offline RagEvaluator retrieval metrics.
Distinguish which fields in mcp-servers-config.json actually flow into the transport/client/gateway runtime, and which are merely governance metadata — to avoid mistaking a field's presence for a capability being in effect.
The AI4J docs site is converging from the old getting-started/ai-basics/guides structure to a canonical structure organized by module and integration path. Describes the migration rules: strong content is not deleted outright, stable conclusions are moved first, legacy notices are added, and the sidebar is the official reading path.
Lists each provider's support matrix for Chat, Responses, Messages, Embedding, Rerank, Audio, Realtime, Image, Video, and Music, based on the AiService implementation.
Pre-launch checklist before integrating AI4J into a real project. It does not require enabling every capability at once, but helps you confirm risk boundaries module by module: minimal module selection, secret management, network isolation, tool exposure, local file boundaries.
AI4J release artifacts, Maven coordinates, and BOM version-alignment strategy, covering each module's role, how to declare dependencies, and the upgrade order.
Maintainer checklist before and after publishing AI4J to Maven Central and GitHub Release, covering version strategy, local verification, and post-release checks.
The four production-grade capabilities layered on the ai4j-agent event stream, explained in depth: node I/O capture and replay (IoCaptureAgentListener/NodeReplayer/NodeIoRecord), resume-cache failure recovery (ResumeCache/ResumableModelClient/ResumableToolExecutor — content-addressed, side effects not replayed), persistent session store, and SHA-256 hash-chain tamper-evident audit (HashChainedEventLog).
Maps AI4J Spring Boot configuration by capability surface via the ai.* prefixes, explaining the flow and layering decisions for single-instance versus multi-instance registry configuration.
AI4J version assessment and pre-upgrade checks: baseline, module compatibility matrix, Java 8 compatibility notes, provider capability differences, and recommended upgrade order.