Tenderix

Tenderix collects every publication from the Swiss public procurement platform and builds a database of what comparable work has sold for. Each morning it sends a digest of supply tenders worth a look. It reads the tender documents, extracts requirements only where it can quote the source, and leaves each go or no-go decision to a person.

You open it at tenderix.karpinity.com and sign in.

1. Collect. Polling ingestion of tenders, awards, corrections and cancellations.
2. Filter. A supply filter narrows the field, and award-price benchmarks decline to quote a median that the sample cannot support.
3. Read. Documents are read in German, French and Italian, with OCR for scans, summaries and translations.
4. Digest. A daily digest goes out at 07:00 Zurich time, by email and Telegram.
5. Decide. Each tender gets a Go, No or Watch. A No must carry a reason tag.

AI that refuses to guess

Requirement extraction records nothing without a verbatim quote from the document. If a model claims a requirement whose quote does not appear in the text, the claim is discarded. Filters fail open when data is missing, and a score that cannot be computed is reported as unavailable.

A model that never leaves the machine

The language model is optional and local. It is called through a command with no outbound network access, so tender text never leaves the machine. Everything still works with no model at all.

Limited reach by design

The client for the procurement platform is restricted to an allowlist of read endpoints. Fetching documents is always a human click.

Tenderix is for suppliers and bid teams who sell to Swiss public bodies. It gives them a morning list of tenders worth a look, the requirements pulled from the documents with quotes, and a record of how past awards were priced. The team still makes the bid decision, with the evidence in front of it.