plus

name	plus
description	Deploy, configure, and operate HAL Plus — the local AI-powered web UI layer for HAL. Use when the user asks to start HAL Plus, switch models, check its health, or understand how it connects to Ollama and HAL MCP.

HAL Plus

HAL Plus is the local web UI for HAL. It runs as a container alongside HAL MCP and proxies queries to an Ollama model on the host. Answers are grounded through the HAL MCP runtime tools.

Lab Assumptions

• Ollama must be running on the host (ollama serve) before hal plus create.
• HAL Plus connects to Ollama via host.containers.internal:11434 from inside the container.
• HAL MCP (hal mcp serve) runs alongside HAL Plus on the same container network.
• Default model preset: qwen3.5 (32k context window).

Primary Commands

hal plus create
hal plus status
hal plus delete

Model Presets

hal plus create --model qwen3.5      # default — balanced, 32k context
hal plus create --model gemma4       # lighter, faster cold starts

Both presets are HAL-managed — no manual ollama pull needed.

Custom Model or Modelfile

hal plus create --model qwen3.5 --keep-alive 5m
hal plus create --model qwen3.5 --model-config ./Modelfile

Custom Image

hal plus create --image localhost/hal-plus:testmiche

Use a locally built image (e.g. during development or testing a branch).

Lab Endpoints

Service	URL
HAL Plus UI	http://hal.localhost:9000
HAL Plus UI (dark mode)	http://hal.localhost:9000/dark
HAL Plus API	http://hal.localhost:9000/api

Architecture Notes

• HAL Plus API server runs on port 9001 internally; HAL forwards port 9000 → 9001.
• HAL MCP is started as a sibling container and reached via HTTP at http://127.0.0.1:18080/mcp from inside HAL Plus.
• Answers are classified by intent route:
  • Route A (knowledge/factual): short prose + one command + one doc
  • Route B (operational): Preflight → Run → Check → Verify → Docs — Qwen wraps the MCP-grounded block in humanized prose
  • Route C (follow-up): prior matched behavior context injected as model grounding
• Status/health checks bypass the Qwen wrapping step and stream the deterministic answer directly.

Edge Cases

• If HAL Plus reports LLM offline, ensure Ollama is running (ollama serve) and the model is present (ollama list).
• If MCP tools are missing, run hal mcp status to confirm the MCP server is reachable.
• If hal plus create fails with an image pull error, either push the image to a registry or use --image localhost/hal-plus:local with a locally built image.
• hal plus delete removes the container but does NOT unload the Ollama model from the host.