24 docs tagged with "integration"

Discover agents, exchange JSON-RPC tasks, stream SSE updates, and expose your ai4j agent as an A2A service with optional auth — JDK stdlib only.

Explains the real wiring of ACP as a headless host for IDEs/desktop shells: newline-delimited JSON-RPC, session lifecycle RPCs, available_commands_update, session/load replay, and the server-side reverse session/request_permission.

Introduces the ai4j-agent Sandbox SPI: how the SandboxProvider/SandboxSession contract hands execution off to an isolated environment, how AgentSessionSandboxBinding keeps only a non-sensitive summary, the three official real providers (Daytona, E2B, CubeSandbox), and how AgentBuilder/CodingAgentBuilder consume the sandbox.

Explains how ai.agentflow.* auto-configures the multi-profile AgentFlow registry (AgentFlowRegistry) under Spring Boot to wire external workflow platforms such as Coze, Dify, and N8N into ai4j, plus the resolution rules for default-name, conditional beans, and the default AgentFlow.

Explains the ai4j-plugin-ask-user sample plugin: it expresses the human confirmation an Agent needs as a host-mediated JSON envelope, contributes the ask_user tool and command/Skill/Prompt resources, and itself opens no UI, reads no stdin, and does not block — the host decides presentation and recovery.

CubeSandboxProvider is AI4J's first real remote sandbox adapter, mapping SandboxProvider/SandboxSession to the CubeSandbox CubeAPI control plane and the envd execution API.

Explains the ai4j-plugin-dynamic-workflow sample plugin: the model writes complex tasks as a deterministic workflow script, the plugin only returns a host-mediated request envelope, and actual execution is optionally taken over by the ai4j-agent runtime; the built-in Nashorn executor disables Java interop by default.

A JDBC persistence solution for promoting FlowGram from a single-process demo to a platform backend, covering the task lifecycle and the task store boundary.

Breaks down the 5 integration depths for third-party MCP in AI4J: single-client direct connection, config-driven gateway, runtime dynamic add/remove, user-level isolation, and the Agent exposure allowlist, with common pitfalls.

How MCP becomes a live tool surface in the Coding Agent: two-layer config (global definition + workspace enablement), five server runtime states, tool name collision checks, and the independent per-session MCP injection path under ACP.

AI4J does not ship a built-in MySQL configuration center, but it reserves the McpConfigSource and McpGatewayConfigSourceBinding extension points; this page covers how to implement a database configuration source and round out naming, secret, and audit governance.

How to configure OpenAI-compatible relay platforms (including TroveBox) in AI4J: plain Java and Spring Boot single/multi-profile setup, endpoint path resolution, and common 401/404 troubleshooting.

The standard approach for building an ingestion, retrieval, and rerank workflow on a Pinecone backend — covers namespace-based knowledge isolation and unified abstraction boundaries.

Explains the Agent lifecycle hooks in ai4j-extension-api: how plugins observe events such as BEFORE_TURN/BEFORE_MODEL_REQUEST/BEFORE_TOOL_CALL/ON_COMPACT, the event payload, exception strategy, and the distinction from Guardrails.

Explains how ai4j-coding routes the Coding Agent's bash exec to a live SandboxSession (P3 first slice), covering the current API, unrouted tools, the relationship with approval, and the security boundary of non-sensitive sandbox summaries.

An enhancement that adds real-time public web search to the answer chain, covering SearXNG configuration, its boundary with RAG, and fallback strategies.

AI4J HTTP stack extension approach: govern production concurrency, connection pools, and network isolation via the DispatcherProvider and ConnectionPoolProvider SPIs.

Deep dive into the real auto-configuration chain of ai4j-spring-boot-starter, its initialization order and conditional configuration boundaries, to understand the unified Configuration and the failure propagation paths.

Explains when to override default Beans along AI4J's abstraction layers in Spring Boot, which layer to choose, and typical override points, so you avoid bypassing the unified container model.

Summarizes the recommended layering and engineering organization patterns for integrating AI4J with Spring Boot, drawing clear boundaries between Web, AI4J calls, and Tools, RAG, and Workflow responsibilities.

A solution for persisting agent sessions to JDBC, covering cross-instance recovery of tool results, runtime state, and compaction summaries.

A multi-turn chat session persistence solution based on Spring Boot + MySQL, covering JdbcChatMemory wiring, sessionId binding, and trimming policies.

Overview of ai4j-spring-boot-starter: when to use it, the minimum integration path, auto-configured capabilities and extension points, plus a pre-launch checklist.

Streamable HTTP transport for AI4J MCP: AUTO profile discovery, modern vs initialization-era peers, protocol headers, publishing a server, and upgrade paths.