Intent filtering
Each watch or tag can carry a plain-text intent — "notify me only when the price drops" or "alert when the item goes out of stock". On every detected change the AI evaluates the diff against it and suppresses irrelevant noise.
Tip: intent evaluation benefits from a capable model — recommended qwen2.5:7b locally, or gpt-4o-mini / gemini-2.0-flash. Very small models (≤3B) may misjudge numeric comparisons.
AI Change Summary
Instead of raw diffs, receive plain-language summaries in notifications — "Price dropped from $89 to $67" or "3 new items added to the listing". Set a global default prompt here, or override per watch or tag.
Minimal cost
The AI sees only a unified diff of what changed — never full page HTML. Low-cost models like gpt-4o-mini or Gemini Flash handle this well, typically fractions of a cent per check.
✓ AI / LLM configured: ollama/qwen2.5:14b
AI Provider
Master switch — when off, all AI lookups are skipped even if a provider is configured below.
Reasoning models (Qwen3, DeepSeek-R1, Gemma 3, etc.) emit chain-of-thought before the final answer. This multiplier scales every max_tokens cap for this endpoint to leave reasoning room. Defaults to 5x; raise it if responses come back truncated or empty, lower it (down to 1x) if you want tighter limits on a paid endpoint. Applied to Ollama and OpenAI-compatible endpoints — other cloud providers (OpenAI, Anthropic, Gemini) keep their original tight caps.
Asks the server to answer without thinking first, which is much faster. Unticked, the server decides (llama.cpp: --reasoning-budget). Works with Ollama, and on OpenAI-compatible servers (llama.cpp, vLLM) with models whose chat template supports enable_thinking (Qwen3 and others).
✓ AI / LLM configured: ollama/qwen2.5:14b
✕ Remove provider
Your API key is stored locally and sent only to your chosen provider. On each detected change, the watch's diff and extracted text are sent to the LLM — no full page HTML.
✕ Clear all summary cache
Removes all cached AI change summaries across all watches. They will be regenerated on the next check.
Enables litellm verbose output (routed through loguru). Useful when diagnosing provider errors or empty responses. Leave off in production — generates a lot of log volume.
Default AI Change Summary
Used for all watches unless overridden by the watch or its tag/group.
Modify default prompt
Behaviour
When enabled, the {{diff}} notification token shows the AI summary instead of the raw diff. Use {{raw_diff}} to always get the original.
When enabled, the AI will be used as a last resort to extract price and stock status from product pages where no structured metadata (JSON-LD, microdata, OpenGraph) is found.
For Gemini 2.5+ models. For Claude Haiku 5.5, Sonnet 5 and Sonnet 5.5, Off switches thinking off and any other value keeps the model default. Thinking tokens improve reasoning quality but count against the output budget. Set to Off if summaries are being cut short.
Upper limit on tokens the AI may use when writing a change summary. Higher values allow longer summaries but cost more.
Token & Cost Tracking
No AI usage recorded yet.
Chrome Webstore