Skip to content

AI summaries (Ollama)

What it does: Summarises notes and feed articles in 3–5 bullet points with a local Ollama (#75). Nothing leaves your home network.

Setup

  1. Run Ollama in your network and pull a small model, e.g. ollama pull llama3.2:3b or qwen2.5:7b.
  2. In Waylight open Settings → AI summaries, enter the address (e.g. http://192.168.1.10:11434) and pick a model from the list. The connection is tested before saving.
  3. A Summarise button then appears on notes and feed items. Until then the feature stays completely hidden.

How it works

  • Notes: the note text is summarised.
  • Feed items: the server loads the article behind the link and keeps the readable text. It uses the same protection as bookmark fetching (no internal addresses, 2 MB limit, timeouts).
  • The input is cut at the limit you set (12,000 characters by default). The summary says when only the beginning was summarised.
  • The summary is written in the UI language (German or English).
  • Summaries are cached per content, model and language, so the same text never costs a second run. A changed note gets a new summary. At most 500 are kept, oldest first out.

Ask the AI in the search (#162)

Type a question after a question mark in the search field, e.g. ? why was jellyfin offline last night, then choose Ask the AI. Nothing runs while you type.

  • The model gets the question plus a compact status text built in your browser from what you can see anyway: host load, offline services, container states and the last timeline events.
  • No settings, credentials or other data are sent.
  • The answer streams in, is marked as AI-generated and can be stopped.
  • Guests and kiosk sessions cannot use it, and the code is only downloaded when you ask.

Privacy and security

  • Home network only. The Ollama address must be a loopback or private IP, a single-label name (ollama), or a local name (.lan, .local, .home.arpa, .internal, .fritz.box …). Public addresses are refused.
  • The connection uses the integration guard (no link-local or metadata addresses). Generating gets up to 3 minutes, enough for a model running on a CPU.
  • Switching the feature off deletes the setup and all cached summaries.

Outbound requests

None until set up. After that: your Ollama server, and the feed article pages you summarise.

API

GET/PUT/DELETE /api/summary/settings, POST /api/summary/models, POST /api/summary. See api/modules/summary.yaml.