Documentation

The complete manual, in two languages

Sixteen chapters and three appendices covering every subsystem, the full class reference, and the reasoning behind each design decision. Written alongside the code, generated from it, and verified against it.

Table of contents

What is inside

1Introduction & OverviewWhat jOpenAgent is, and what it is not
2Getting StartedSetup, project settings, first agent
3Core ConceptsAgents as objects, annotations, the trampoline
4Generation StrategiesPREDICT, CODE_ACT, TOOL_CALLING, REFLEXION
5Tools, MCP & Structured CapabilitiesSchemas from signatures, MCP over stdio
6MemoryStores, embedders, recall weighting
7SkillsSKILL.md bundles, scripts, the registry
8Conversation History & SummarizationToken budgets and the summariser prompt
9Context BlocksStatic and dynamic blocks
10Multimodal ContentImage, audio, video, attachments
11Tracing & ObservabilitySpans, four exporters, three viewers
12Code-as-Action SandboxJShell, the validator, out-of-process mode
13Evaluation HarnessSuites, cases, tiers, scorers, experiments
14LLM ClientsAnthropic, OpenAI, local servers, embeddings
15Tutorial WalkthroughAll nineteen examples, explained in order
16Safety NotesWhat the sandbox does and does not guarantee

Other resources

Beyond the manual

Quickstart

Eight steps from an empty project to a traced, typed, tool-using agent. The fastest path if you learn by doing.

Open the quickstart →

The 19 examples

Runnable classes, one new idea each, mirrored by chapter 15 of the manual. Read them in order or jump to the one you need.

Browse the examples →

Javadoc

Every public type is documented in source. Build it with your usual javadoc run, or read it inline in your IDE. There is no separate doc site to fall out of date.

Get the sources jar →

Start reading, then start building