Fourteen days of the graph tier, no card.

Four things, all of them checkable on your own content during the trial. An answer can be assembled from several documents rather than summarised from one. A search reaches relevant material that shares no wording with the question. Every claim names every document behind it, not one representative mention. And the vocabulary your content is read against is versioned, previewable and revertible, so it can change as your corpus does.

Nothing is deleted and nothing is switched off. The organisation settles onto the free tier, and any graph workspace you built during the trial keeps working — the payment gate is on creating a graph workspace, not on using one. What changes is capacity: you are on the free quotas from then on, and creating a further graph workspace needs a card. No card is taken automatically, because nothing on this site charges yet.

Because creating an organisation is how anyone reaches the product at all, and a card wall there would stop evaluation before it starts. A new organisation's first workspace is always RAG, and RAG's model spend is small — embeddings when a document arrives and one answer call per search. The graph tier also spends model calls reading every document as it arrives, which is why that one needs a payment method once the trial ends.

The tenancy boundary. A workspace scopes all of your data and all of your API keys — documents, entities, the graph, the ontology, everything. An organisation can hold several, with the number set by your plan (the table above), and the GRaaS entitlement is granted at the organisation level — so one payment method unlocks the graph tier for every workspace you own.

They stay indexed and searchable, but they are not retroactively extracted into the graph. Upgrading changes what happens to documents ingested from that point on. If you want your existing corpus in the graph, re-POST it — ingest is an idempotent upsert on the document's external ID, so re-sending is safe and replaces rather than duplicates.

A workspace stays on the tier it was created with, so the two remain comparable rather than drifting into a state nobody can reason about. To compare them, the search history in the console can run any past query against RAG-only retrieval and show you both results side by side — that needs nothing extra. For a standing comparison, keep one workspace of each tier on the same content, which needs Starter or above since the free tier holds a single workspace.

Every limit is an organisation-wide pool, summed across all of your workspaces — payment is at the organisation level, so the quota is too. Document limits count documents currently stored, not documents ever sent, so deleting a document frees its slot. The monthly extraction and search limits are a rolling 30-day window, and the per-minute search rate is a burst ceiling that protects shared infrastructure.

It depends on the limit, and not every one is enforced yet. A search over the per-minute limit, or an API search over the monthly one, is refused with a 429, and creating a workspace past your plan's count is refused. A document past the document limit is accepted and held — stored but not processed or searchable — until you free a slot or upgrade, rather than refused. A graph document past the monthly extraction limit is searchable straight away, and its graph is built once the allowance refills. Extractions beyond the concurrent limit wait their turn. The daily ingest and per-batch figures are displayed but not yet enforced.

Not yet, and we would rather say so than imply otherwise. Our security page describes exactly what is in place today — workspace isolation, scoped keys with rotation, single sign-on through WorkOS, session revocation — and what is still on the roadmap.

Yes. Raise it during the demo conversation and we will get it in front of you before you send any production content.

No, and you currently cannot. Extraction, embedding and answer synthesis run on models we select and pay for, which is part of what the GRaaS tier covers. If a specific model or region is a requirement for you, that is an Enterprise conversation.