Examples

Task-first index of how-tos, ordered by the adoption funnel: try it in a terminal, then over HTTP, then wire it into your app. Each page is copy-paste-runnable. No tutorial content lives here — this page is just the router.

Run the demo app first

Every HTTP example below talks to the demo app on http://localhost:8080:

./gradlew :app:bootRun

The port comes up immediately but the model loads asynchronously — wait for state: READY in the logs (or poll /actuator/health) before sending chat requests. Until then the endpoint returns 503 { "error": "model loading" }. The CLI examples need no server.

Try it (zero install)

• CLI — npx @deemwario/mochallama chat -m qwen2.5-1.5b. Proves the local engine from a terminal with no Java install (ships a jlink JDK-22 image). 0.1.6 adds real multi-turn chat with persistent sessions: --resume <id>, mochallama sessions, in-REPL /reset /help /exit, --no-save.

Drive it over HTTP

• curl — POST /v1/chat/completions (+ SSE stream:true) and GET /v1/models straight from the shell. Proves the OpenAI-compatible wire API.
• OpenAI SDK (Python) — the official openai client with base_url pointed at the local endpoint. Proves drop-in compatibility — point existing OpenAI code at localhost with one line.

Embed it in your app

• Spring Boot — add the mochallama-spring-boot-starter; autoconfig exposes /v1/chat/completions (+SSE), /v1/models, and Actuator health/metrics. The same page covers the Spring AI adapter (mochallama-spring-ai): add it alongside the starter and inject a ChatClient / ChatModel. Proves in-process, tool-calling inference inside your own JVM process — no sidecar.

Tool calling & streaming

• Tools & Streaming — the full tool-calling round-trip plus SSE consumption. Proves that tools, tool_choice, and streaming work together (mochallama is tool-calling-only; non-tool models are rejected at load with MODEL_NOT_TOOL_CAPABLE).

Reference

• Switching models — presets (qwen2.5-1.5b default, qwen2.5-3b, qwen3-4b, phi-4-mini) and any tool-capable Hugging Face GGUF id: see Model Profiles.
• Metrics & health — inference meters and the readiness indicator: see Metrics.

Coordinates

Maven Central io.github.deemwario:*:0.1.6 · npm @deemwario/mochallama · requires JDK 22+ (FFM GA).