Glossary

The domain terms used across the Infinia docs, each with a one-paragraph definition and a pointer to where it is explained in depth.

FileRef

An opaque handle the host mints to let a sandboxed plugin refer to a file or directory without ever seeing a real path. Its shape is {id, name, kind, access, size}, where id begins with ref_. The plugin UI passes the FileRef straight into an RPC; the host's PluginProcessManager rewrites it to an absolute filesystem path before dispatch, so the worker receives a real path it can open. Grants live in memory only and do not survive a host restart. See File I/O.

MF (micro-frontend)

A self-contained UI bundle (the plugin's ui/ directory) that the host serves as static assets under /plugin-runtime/{id}/** and loads exclusively into a sandboxed iframe under a strict Content Security Policy. Inside the iframe, @infinia/plugin-sdk's FengYuClient bridges to the host over the shared, versioned postMessage protocol. See UI Micro-frontend.

SETUP mode / APP mode

The two runtime modes the backend auto-selects at launch, based on the datasource config at <program-working-directory>/.fengyu/config/datasource.properties:

• SETUP mode boots SetupApplication with no JPA and serves the first-launch wizard under /api/setup/* (token-bypassed). Used before any database is configured.
• APP mode boots FengYuApplication with fengyu.mode=app and the full persistence + AI + plugin stack.

If the config exists but the DB is unreachable, the launcher backs the config up to a .bak sibling and falls back to SETUP mode. See Backend — SETUP vs APP mode.

Sidecar

The worker process a plugin ships — typically a shaded backend/worker.jar launched by java -jar. It runs as its own OS process, spawned and owned by the host's PluginProcessManager, and speaks JSON-RPC 2.0 over stdio. Because it is out-of-process, a worker crash or hang cannot take down the host, and the worker cannot reach into host beans or the JPA session. "Sidecar" emphasizes that it lives alongside the host, not inside it. See Worker (JSON-RPC).

Virtual user

The local identity Infinia creates on first APP-mode startup: id 1, name ZFlow-Summer, role admin/local. Conversations and other user-scoped records attach to this identity. There is no password or login flow — the virtual user is implicit on a single-user install. See Database — Virtual user.

JSON-RPC worker

The protocol and process model for plugin backends. The worker reads newline-delimited JSON-RPC 2.0 requests from stdin ({jsonrpc:"2.0", id, method, params}), dispatches to registered handlers, and writes one response per line to stdout (result or error). stderr is reserved for logs. The toolchain/sdk-java artifact ships JsonRpcWorker, a tiny runtime with .on(method, handler) registration. See Worker (JSON-RPC).

.fyp

The plugin package format — a zip archive with three parts: manifest.json (metadata, permissions, AI tools), ui/ (the micro-frontend assets), and backend/worker.jar (the sidecar executable). Installed via the marketplace (POST /api/plugin-market/upload for a .fyp, or upload-native for a local path). Built with fengyu build. See Plugin Overview and Build & Deploy.

uiEntry

The resolved UI entry URL on an InstalledPluginDescriptor — the page the host opens in the plugin's sandboxed iframe. It is derived from the manifest's ui.entry (typically ui/index.html) and served under /plugin-runtime/{id}/<entry>. See Plugin System — Installed plugin descriptor.

ToolCallback

The Spring AI abstraction for a tool the model can call. Infinia aggregates every built-in @FengYuTool and every enabled plugin's declared aiTools into a single ToolCallback[], so plugin tools and built-in tools are indistinguishable on the wire — both surface in GET /api/agent/tools and both can be invoked from chat or agent runs. See AI Tools.

MD3

Material Design 3 — the design system implemented by the host UI (Vuetify) and @infinia/plugin-ui. Plugin UIs render inside their own iframe realm and track host theme/environment changes through the SDK bridge. The purple #6750A4 theme color is the MD3 baseline. See Design System and UI Micro-frontend.

Ollama backend

The local AI backend mode. It talks to an external ollama serve process over Ollama's local HTTP API — the backend does not load a GGUF model in-process. To use it, run ollama serve separately and select local in AI config. The other modes are openai, anthropic, and deepseek (OpenAI-compatible). See AI Chat — Backends.

Next steps

• REST API — where these terms show up in endpoints.
• Architecture — Plugin System — how FileRefs, MFs, and workers connect.
• Architecture — Backend — SETUP/APP mode and the sidecar process model.