It can. ReceiptDB stores vectors in PostgreSQL with pgvector and serves search itself, with your policies enforced. It plugs into LangChain and LlamaIndex as a vector store, so adopting it is a small code change rather than a migration project. One command imports a collection from Pinecone, Weaviate, Qdrant, Chroma, Elasticsearch, OpenSearch or plain pgvector, labelled as unknown provenance until you say otherwise, with each point's readers and data subjects taken from its metadata. Lineage and labels add 1 to 9 ms per search. Vectors are stored at half precision, so the search index is 40% smaller than plain pgvector's 32-bit one (1.2 GB against 2.0 GB at a million chunks) and builds and searches as fast, for 99.6% of the true top ten at a million chunks against 32-bit's 100%, and the same recall on real embeddings.

With Enterprise's connectors following Google Drive or SharePoint, about a minute after the source reports the change: the document and everything built from it leave search and answers for the people who lost access. For documents where a minute is too long, the live check reads permissions at the source again before a document is returned, and leaves it out if the source can't answer.

Yes, if you turn on the detector, which runs in your own environment. Its built-in engine finds email addresses, phone numbers, card numbers and IBANs, and, when you name them, US and UK national identifiers such as National Insurance numbers, with no model and no extra service; Microsoft Presidio adds names. It flags personal data a writer didn't declare and proposes the people it can identify, such as an email address, for a rule or a reviewer to confirm. It records entity types and scores, never the text it matched, and it only adds restrictions. Like any detector it can miss things, so the labels your writers declare still come first.

No. ReceiptDB runs on PostgreSQL with the pgvector extension, self-managed or managed, including Amazon RDS. The lineage graph and audit log are ordinary tables, so your existing backups, monitoring and access controls apply.

Any. Answers, grounding checks, agent memory and the injection scanner run on Anthropic, any OpenAI-compatible endpoint including local models, or AWS Bedrock, and the MCP server lets the agents you already run record their answers with receipts. Content labelled for local use never goes to a hosted model.

Nothing is held hostage. After expiry there's a grace period; after that, only new Enterprise features stop. Reads, decryption, erasure and access to every tenant keep working, through restarts too, so your data and your obligations to the people in it never depend on a renewal.

The Community edition is under the Functional Source License (FSL-1.1-ALv2). Use, modify and redistribute it for any purpose except offering a competing product or service; each release becomes Apache-2.0 two years after it ships. Enterprise is a separate commercial licence.

Pricing depends on your deployment. Talk to us for a quote.