Specs against clinical need
Item specs checked against formularies, pharmacopoeia standards and regulator registrations — so the line item is what clinicians actually prescribe, not a supplier's rebrand of last year's catalogue.
Agentic tender review for public health procurement
Quorum reads a health tender end to end — specs against clinical need, quantities against real consumption, prices against other buyers. Every finding cited. Then it stops and hands over: the committee decides.
NO TOOL CALLS LOGGED YET — START A RUN TO SEE ACTIVITY
EVERY TOOL CALL LOGGED — INPUTS, OUTPUTS, TIMESTAMPS
Sourced live from 22 datasets across 9 national and county portals.
Overview
Sorting bids, ticking off documents, checking each bidder, comparing prices and writing the report is committee work, done by hand. Quorum does the same work in a single review run — and then stops.
AUDIT TRAIL
Every tool call logged — inputs, outputs, timestamps.
ONE TENDER, END TO END
FIG. 01 — THE CLOCK
COMMITTEE WORK, BY HAND
26days
COMMITTEE WORK, BY HAND
~20min
WHAT YOUR AGENT SHOULD TAKE
MEASURED, NOT CLAIMED — on the record's staged packs, the shipped review graph (files → requirements → cited findings → human gate) runs in 4–9 ms deterministic, 20–28 s on the configured model. The read-and-flag pipeline scales to volume measured at ~0.9 s per MB of pack text. The ~20 min stays the envelope for full real-world packs — still two orders of magnitude inside 26 days, and the gavel stays human either way.
↑ THE FIVE JOBS ABOVE ARE THE 26 DAYS · THE RED SLICE IS THE ~20 MIN ↑
COMMITTEE DECIDES
Quorum stops here and hands over a sourced report.
One rule — not negotiable
AGENT PREPARES→COMMITTEE AWARDS
Every finding cites the document it came from
An unsourced flag against a supplier is worse than no flag
Procurement law across Africa keeps the award with a human committee
The theme — health & life sciences
Medicines, devices and commodities are the largest, most error-prone lines of public health spending. One theme, mastered — depth beats breadth when the audit comes.
Item specs checked against formularies, pharmacopoeia standards and regulator registrations — so the line item is what clinicians actually prescribe, not a supplier's rebrand of last year's catalogue.
Order sizes reconciled with consumption and stock levels at the facilities the goods are destined for — buying this year's caseload, not last year's budget line.
Unit prices benchmarked against the same commodities bought by other public buyers on the same OCDS portals — the outlier is visible before the award, not in an audit two years later.
ONE THEME SHIPS AT A TIME — THE SAME WORKFLOW, QUEUED NEXT
Reel 01 — Generated cinema
A two-minute film in six chapters — every frame generated by an open-weights model, cut to the same visual system as the product. The story of a run: the stack, the read, the evidence, the flag, the gate, the award.
QUORUM ORIGINALS
THE TWENTY-MINUTE
TENDER.
A FILM IN SIX CHAPTERS · SHOT ENTIRELY BY MODELS
PLAY THE FILM · ≈ 01:57
EVERY FRAME MODEL-GENERATED · NO CAMERA, NO CREW, NO STOCK · CUT IN THE OPEN
Inside the run
Nothing below is an enterprise upsell — it ships in every run.
LIST 01
Reads the whole tender. Notices, line items and awards straight from OCDS portals — no scraping, no re-keying.
Runs tools that do real work. Document checklists, bidder verification, price comparison — logged MCP tools, not read-only wrappers.
Borrows what is proven. Community and official MCP servers where they beat hand-rolled code — and says so in the log.
Runs on open weights. Procurement data often cannot leave the country. Open-weights models by default; frontier alongside if you allow it.
Logs every action. Inputs, outputs, timestamps — and a named human gate on every irreversible step.
Stops before the award. The file is prepared; the decision stays with the committee. Always.
LIST 02
A sourced report findings with citations down to the page — ready for the committee pack.
A flag register, missing documents, lapsed registrations, price outliers — each with its evidence attached.
A full audit log every tool call with inputs, outputs and timestamps. Exportable for auditors and oversight bodies.
A one-page architecture brief, which tools run, which models, and exactly where your data lives.
An eval harness eight-plus test tasks with pass and fail results — trust the run before you scale it.
The output
One document, three panes. The report the committee reads, the flag register it rules on, and the audit log that proves how every line got there.
SOURCED REPORT
NO RUN YET · EVERY CLAIM CITED
LOADING FROM THE QUORUM AGENT…
QUORUM
THE COMMITTEE SIGNS — THE AGENT NEVER DOES
Live from this workspace
Counted from the review runs on this server. Zero means no run has happened yet.
Pricing
Every tier keeps the two non-negotiables: every finding cited, a human gate on every award. No exceptions.
ENTERPRISE
Custom
ON-PREM, OPEN WEIGHTS, YOUR JURISDICTION
For ministries and regulators with data-residency and integration mandates.
Getting started
Your first run reads the tender end to end, on a real portal, with real notices.
Already have access?
FIRST-RUN READINESS
0 of 5 ready
TICK ALL FIVE — THEN LAUNCH YOUR FIRST RUN
REQUEST A PILOT
Three tenders, free — on your own portals.
NO CARD · NO DECK · A REAL RUN