OwnedPulse — regulatory intelligence with a source for every answer

What it does

Five things, each visible in the interface. Nothing happens in the background that you cannot open and read.

Monitors your sources

Guidance, guidelines and Q&A documents from EMA, FDA and ICH, plus the feeds you add. Each document is ingested with a record of when, from where, and in what version. Superseded documents stay in the registry and are marked as such.

Answers with citations

Ask a question; the answer is composed only from retrieved passages and every claim carries a citation. Passages that were retrieved but not used are listed as such, so you can see what the system read and chose not to rely on.

Refuses when it should

When the corpus does not contain the answer, the response says so. No filler sentence, no plausible guess.

Opens the trace

One click opens the full record behind an answer: the model call (model, tokens, latency), the query parameters, each cited passage with clause, page and score, and the ingestion record of the document it came from.

Exports the evidence

Any answer exports as a PDF with query identifier, timestamp, the version manifest of everything involved, and the referenced documents — a record you can file with a decision.

In the interface

Screens from the current release, running against an EMA, FDA and ICH corpus.

Query page. Ask in plain language; filter by regulator, document type and date before you ask.

Traceability drawer. Model call, query parameters, cited passages with clause, page and score, ingestion record.

Audit export. The answer as a filed record: query, cited answer, export ID and timestamp, over 23 pages of evidence.

Corpus registry. Every ingested document with agency, type, publication date and chunk count; filter by body, type and date.

Six-minute demo

One question, followed end to end: the answer, the traceability drawer, the audit export, and the corpus registry. Narrated; no sales language.

The video is hosted on YouTube. Opening it leaves this site.

Watch on YouTube

Under the hood

Open-weight model served by Ollama. Qdrant for retrieval, PostgreSQL for the document registry and chunk archive, Langfuse for call-level traces. Docker Compose on a single Linux server with one GPU. Every component runs inside your network; the only external traffic is fetching the regulatory sources you configured.

What it is not

Not a document management system. It reads regulatory sources; it does not manage your controlled documents.
Not a replacement for Veeva Vault, Documentum or your QMS.
Not a validated GxP system. It is an open-source demonstrator; validation, if you need it, happens in your quality system with your evidence.
Not for GMP-critical decisions. It supports reading and finding; the decision stays with a qualified person.

Get it

Install it yourself from the repository, or have it deployed and configured for your sources on your infrastructure, with thirty days of support.

Deployment service

Source code github.com/ownedai/ownedpulse Demo video youtu.be/b8GvY9ZtY5Y The rules it is built on ownedai.dev/approach