Free vs AI-enhanced

bAInquet has a free tier that produces deterministic, fully reproducible data with zero LLM work, and paid tiers that add AI enrichment on top. Every value is provenance-tagged so a consumer can always tell deterministic data from AI-derived data.

The free tier: deterministic data

The free tier triggers zero LLM work anywhere in the pipeline. A free-tier job that attempts LLM work is a contract violation rejected at enqueue. The result is data that is:

• Reproducible. Given identical graph content and a frozen build timestamp, exporter output is byte-identical.
• Auditable. Every fact records the deterministic extractor that produced it and a confidence score.
• Free of model risk. No hallucination, no inference, no model cost.

What the free tier includes

The four deterministic extractors run in priority order, each stamping a provenance method:

Provenance method	Confidence	Source of the value
cms_field	0.98	Structured CMS fields supplied in the item json bag
schema_org	0.95	schema.org JSON-LD on the page
seo_meta	0.85	OpenGraph / SEO meta tags
text_extraction	0.75	Deterministic extraction from plain text

The free tier publishes these node files (all produced by free exporters):

manifest.json, ai.json, llms.txt, entities.jsonl, facts.jsonl (and per-language facts.<lang>.jsonl), relationships.jsonl, chunks.jsonl, qa.jsonl, sources.jsonl, schema.jsonld, and sitemap-ai.xml. See Node files.

The paid tiers: AI-enhanced data

Paid (advanced) tiers add LLM enrichment on top of the deterministic base. Enrichment is opt-in, budgeted, and never blocks the deterministic path. It can include:

• Gap-fill: inferring values the deterministic extractors did not find.
• Inferred relationships: edges suggested by a model rather than read from structured data.
• Grounded Q&A: synthetic question/answer pairs generated from the facts.
• Tuned chunks: retrieval chunks shaped for better model consumption.

Advanced tiers also gate the trust.json and openapi.json exporters, which run only when plan-derived capabilities allow.

Every enriched value is tagged

This is the contract that makes the paid tier safe to consume: every model-derived value carries the provenance method ai_enhanced with a lower confidence (0.60 to 0.90), and the merge policy ranks it below every deterministic method (cms_field > schema_org > seo_meta > text_extraction > model-inferred). A consumer can filter on provenance to use deterministic facts only, or to weight AI-derived facts lower.

Provenance tagging: cms_field vs ai_enhanced

	cms_field (and other deterministic methods)	ai_enhanced
Produced by	Deterministic extractors, no LLM	An LLM, advanced tier only
Confidence	0.75 to 0.98	0.60 to 0.90
Reproducible	Yes, byte-identical	No (model output varies)
Merge precedence	Higher	Lowest
Available on free tier	Yes	No

Because the tag travels with every published fact, a downstream AI consumer reading the node can separate owner-supplied deterministic data from AI-derived data at consumption time. See Core concepts.

Comparison

	Free	Advanced (paid)
Deterministic extraction (cms_field, schema_org, seo_meta, text_extraction)	yes	yes
LLM gap-fill, inferred relationships, grounded Q&A, tuned chunks	no	yes
trust.json, openapi.json exporters	no	yes
Reproducible, byte-identical output	yes	the deterministic base; enriched values vary
Provenance tagging on every value	yes	yes (ai_enhanced on enriched values)

Billing status

Plan entitlements and usage are readable at GET /v1/organizations/:id/entitlements and GET /v1/organizations/:id/usage (plan-derived constants). Live payments are not wired in the current build: the Stripe integration is scaffold-only, with no checkout, portal, or webhook routes.

Pricing

For current plans and pricing, see the marketing site at bainquet.online.

Related

• Core concepts: the deterministic-first principle.
• Node files: which exporters run per tier.
• Organizations and websites: the entitlements endpoint.