Documentation Map
This page defines the canonical reading map of the AI4J documentation. It does not replace the feature pages — it answers two questions:
- What question does each of the eight top-level sections answer, and which one should I enter?
- What is the single entry point for a given capability, so you don't bounce between pages that cover similar ground?
The eight top-level sections
The docs are organized by reader intent, not by code module. The top-level order is the recommended reading order:
| Section | What it answers | Entry |
|---|---|---|
| Getting Started | What AI4J is, whether it fits, and how to run your first snippet | Intro |
| Capabilities | What works without the Agent module: models, media, tools, skills, chat memory, RAG, MCP | Capabilities Overview |
| Agent | How to build autonomous agents: runtimes, memory & compaction, orchestration, governance, observability & interop | Agent Overview |
| Extending AI4J | How to extend AI4J itself: the jar plugin system and code-level extension points | Extending Overview |
| Products | Products built on the SDK: Coding Agent (CLI) and FlowGram | Coding Agent · FlowGram |
| Integrations | Wiring AI4J into your stack: Spring Boot and recipes | Spring Boot · Recipes |
| Production | Pre- and post-launch checks: security, checklist, troubleshooting | Production Checklist |
| Reference | Specs and background: API, versions & migration, maps, about | API |
The dividing line between Capabilities and Agent is: does this capability exist without importing the ai4j-agent module? If yes, it belongs to Capabilities (e.g. chat memory); if no, it belongs to Agent (e.g. agent memory & compaction).
Single entry point per capability
| Capability you are looking for | Start here |
|---|---|
| Chat / Responses / Messages / Streaming / Multimodal | Models |
| Image / Audio / Video / Music / Realtime | Media Generation |
| Function tools / allowlist / execution model | Tools |
| Skills (SKILL.md, discovery, activation) | Skills |
| MCP (client / transport / gateway / server) | MCP |
| Chat memory (session history) | Chat Memory |
| RAG / vector stores / ingestion / evaluation | RAG |
| Agent runtimes / subagents / teams | Agent |
| Agent memory compaction / context management | Memory & Compaction |
| Approval / interceptors / sandbox | Governance |
| Trace / replay & recovery / A2A | Observability & Interop |
| Plugin packages (jar / SPI) | Plugin System |
| Provider / service / HTTP stack extension | Code-Level Extension Points |
| Coding Agent CLI / TUI / ACP | Coding Agent |
| FlowGram nodes / task API | FlowGram |
| Spring Boot auto-configuration | Spring Boot |
About old links
The historical directory layout (top-level start-here/, core-sdk/, mcp/, etc.) has been folded into the eight sections above. Every old URL redirects automatically to its new location — no manual mapping needed, and old addresses indexed by search engines keep working.
New documentation must land inside one of the eight sections' existing topics; do not open a new top-level directory for a single page.
If you don't know which track to read yet, go back to Choose Your Path.