Settings & AI

Everything here is per-workspace. The most consequential section is AI — and the most important thing to know is that you do not need it.

Alert thresholds. The minimum and the cooldown are what keep a quiet store from tripping an alert on three reviews.

Workspace settings. Without an AI provider the built-in engine does the scoring and everything still works.

AI analysis

Vantage is fully functional with no AI provider. With no key configured, an analysis engine running on your own server scores every review's sentiment, tags themes, builds the SWOT and writes the report summary. All charts, all alerts, all trends work. Nothing is locked.

What adding a key changes

	Without a key	With a key
Sentiment scoring	Yes	Yes, more nuanced
Theme tagging	Yes	Yes, more accurate
Statistics, charts, histograms	Yes	Identical — always computed from your data
Alerts	Yes	Identical
Report summary	Factual, templated	Written analysis in prose
SWOT	Assembled from the data	Reasoned about your specific position
Recommendations	Matched to complaint themes	Tailored to what reviewers actually said

Connecting a provider

Vantage supports Anthropic (Claude) and any OpenAI-compatible endpoint — OpenAI itself, Azure OpenAI, OpenRouter, LiteLLM, vLLM or Ollama.

Your key is stored encrypted, used only from your own server, and billed to your own provider account. Vantage never proxies your requests and never sees your key.

Use Test AI connection after saving to confirm it works.

Two models, two jobs

Vantage uses two models against the same key, because the two tasks are not alike:

Setting	Job	How often it runs	Choose
Scoring model	Scores individual reviews	Once per review — thousands of times	The cheapest model. This dominates your bill and the task is simple.
Report model	Writes the weekly report	Once per store per week	The best model you're willing to pay for. It runs rarely and this is the part you read.

The defaults already reflect this. Leave them alone unless you have a reason.

What it costs

For a store getting 500 reviews a week, with the default model split:

Provider	Per week	Per year
Claude (Haiku scoring + Opus reports)	~$0.30	~$16
OpenAI (mini scoring + GPT-4.1 reports)	~$0.12	~$6

Costs stay roughly flat as review volume grows, because reports are built from a representative sample rather than every review. Actual figures depend on how long your reviews are.

The token cap

Monthly token cap is a hard stop, checked before every request. When it is reached, Vantage falls back to the offline engine rather than spending more. Set it to 0 for no limit.

Current usage and approximate spend are shown at the top of the Settings page.

Collection etiquette

How politely Vantage reads review pages. These settings can only be made more conservative than the shipped defaults — you cannot remove the delay or disable robots.txt compliance.

Setting	Default	Notes
Respect robots.txt	On	Locked on in most installations
Delay between requests	6 seconds	Minimum 4. Random jitter is added on top. Raise this if a site starts blocking you.
Max pages per product	10	Each page is roughly ten reviews. Lower it to collect less on very large listings.
Retries on failure	3	With increasing waits between attempts
User agent	From your `.env`	Identify yourself honestly, with a contact URL
Proxy URL	Empty	Optional. Only needed if your server's IP is being blocked.

Slower is better. There is no prize for collecting quickly. Reviews accumulate over days; a collection that takes ten minutes instead of two costs you nothing and substantially reduces the chance of being blocked.

Alerts

Covered in detail on the Alerts page.

Regional

Timezone determines where your reporting weeks start and end. Set it to your business's timezone so "last week" means what you expect.