Licensing
Your search returns an answer. WINNEX proves it was the complete one — and puts that proof in a signed record an auditor can verify without trusting you. Published pricing, four RAG packages, six components. Start where your pipeline actually breaks.
1. RAG packages
Four packages, each resolving one stage of the RAG chain. Most teams start with Search Proof and add ingest validation or the audit trail as the deployment grows. Bundled components cost less than buying them apart.
| Package | Components | What it solves | Mid-tier / yr |
|---|---|---|---|
| Search Proof | madhava | You have a pipeline. You need proof of what the search excluded. | $120,000 |
| RAG Ingest + Search | normalize + madhava | Bad corpora are burning you before anything reaches the index. | $153,000 -15% |
| Audited RAG | madhava + tracer | You must prove later that the record was neither altered nor removed. | $306,000 -15% |
| Full RAG Proof | normalize + madhava + tracer | The whole chain: input validated, exclusion proven, record sealed. | $315,000 -25% |
Mid-tier shown. Small and Large scale from the component table above. tracer inside a package covers one domain.
2. Components, individually
| Component | Question it answers | Small | Mid | Large |
|---|---|---|---|---|
| winnex-ai-normalize | Is the corpus usable before it reaches the index? | $24,000 | $60,000 | $150,000 |
| winnex-madhava | What did the search exclude, and is the exclusion provable? | $45,000 | $120,000 | $320,000 |
| winnex-tracer | Can that record be rewritten, or deleted? | $80,000 | $240,000 | $520,000 |
Small up to $10M revenue · Mid $10M–$100M · Large above $100M. A licence is per organisation, not per seat.
For a RAG platform, the licence pays back when it protects contracts that already exist.
3. Tracer, by domain
tracer is not one product. A court of accounts, a health insurer and an e-Discovery team have different budgets, different auditors and different questions. Each domain is licensed separately so you pay for the one you need.
| Domain | Buyer | Market anchor | Small | Mid | Large |
|---|---|---|---|---|---|
| tracer-med | Health insurers, hospitals, EHR vendors | Denied claims — 1% avoided pays the licence | $80,000 | $240,000 | $520,000 |
| tracer-jus | e-Discovery, corporate legal, LegalTech | Relativity: mid six figures per year | $110,000 | $320,000 | $750,000 |
| tracer-gap | Banks, insurers, financial reconciliation | Splunk / Vanta-class: $60k–250k/yr | $95,000 | $280,000 | $650,000 |
| tracer-gov | Courts of accounts, controladorias, public prosecutors | Additional Use Grant | Free | Free | Free |
Every domain above has a different break-even. Change the inputs to your own numbers.
4. Implementation audit & certification seal
A licence proves the software is the right one. It does not prove the deployment is. The implementation audit is what turns a correct product into a defensible system — and it frequently pays for itself by finding spend that was going to the wrong place.
| Audit | What is verified | Duration | Price |
|---|---|---|---|
| Readiness review | Corpus validated through the quality gate; embedding provider and dimension confirmed; routing checked against your data. Produces a written go/no-go. | 5 days | $18,000 |
| Implementation audit | Full deployment: retrieval completeness measured against your own ground truth, bound coverage per query class, WORM chain verified end to end, signature round-trip tested, and the signed record reproduced independently. | 3 weeks | $65,000 |
| Certification seal | The audit above plus a published attestation — a verification page a third party can query, and a seal you may display. Valid 12 months, re-verified annually. | 3 weeks | $95,000 |
| Annual re-verification | Re-run of the certification against the current deployment, to keep the seal live. | 1 week | $32,000 |
- What the seal means. The stack was implemented to the standard and verified against your corpus: the quality gate passed, retrieval completeness was measured, the audit trail was confirmed tamper-evident, and the signed records reproduce independently.
- What it does not mean. It is not a regulatory certification, not a legal opinion, and not a statement about your encoder's semantic quality. It is evidence about the implementation.
- Why it pays. Most deployments over-provision embedding and vector infrastructure before measuring whether the corpus is even indexable. The readiness review alone regularly removes spend; the audit then keeps the remainder justified.
The quality gate (winnex-ai-normalize) does part of this automatically at ingest and is included in every licence. The audit is the independent verification of the result — necessary when a third party will rely on it.
5. Assisted implementation mentoring
For teams starting from zero, or rebuilding a process that never worked. Mentoring is not a licence and not an audit — it is guidance while you build, so that the first deployment is one that can later be certified.
| Programme | What we do | Shape | Price |
|---|---|---|---|
| AI stack selection | Model and provider evaluation against your data: embedding quality measured, dimension and routing decided, cost per million queries compared. Ends with a written architecture decision record. | 4 weeks | $28,000 |
| Build alongside | Weekly working sessions while your team implements: retrieval design, the quality gate at ingest, the audit trail, and the failure modes that bite in production. We review code, not write it. | 8 weeks | $64,000 |
| Process and team | How the process changes around the AI: who owns the corpus, who signs a retrieval, what a reviewer checks, how a new model enters the pipeline. Includes role definitions and a runbook. | 6 weeks | $48,000 |
| Full programme | All three, sequenced: selection, build, then process. The path from nothing to a deployment ready for certification. | 18 weeks | $128,000 |
- Who this is for. Companies with no AI capability yet; teams whose RAG pilot produced confident wrong answers; organisations replacing an approximate index because a regulator asked a question they could not answer.
- Why before the audit. An audit tells you what is wrong. Mentoring is what avoids paying for that twice — it is cheaper to build to the standard than to rebuild to it.
- Independence. If we mentored the build, the certification seal for that deployment is issued by a partner auditor, not by us. We state this in the proposal.
Mentoring does not include the licence — components and packages are licensed separately. Many teams pair the Full programme with Search Proof and add the audit trail once the retrieval is stable.
6. Outside the packages
The published tiers cover organisations up to the Large band. Three cases sit outside them — not because they are excluded, but because a single number would misprice them:
| Situation | Why it is not in the packages |
|---|---|
| Big Tech and hyperscalers | Companies of that scale extract a different order of value from the same technology, and no published tier would be right for them. Terms are negotiated individually. |
| OEM at any size | When the licensee embeds the stack in a product it resells, the price follows what it extracts rather than what it earns. Terms are negotiated individually. |
| Asset acquisition | Assignment of IP, code, trademark, DOIs and contracts. Terms are negotiated individually. |
7. Request a licence
Choose the licence you need. The form builds an e-mail with the right fields for that modality — no account, no sales call required.
Choose the licence
Select what you need. The form builds an e-mail with the right fields for the chosen modality — no account, no waiting for a sales call.
8. What the licence does not include
- The audit record is technical evidence, not a certification. It sits underneath a notified body's assessment — it does not replace it.
- The proof covers the vector space, the encoder quality is yours to control. A poor embedding yields a correct proof and an irrelevant answer; the quality gate measures and reports this before indexing.
- Latency trades against the proof. 20 to 100× slower than HNSW at equal recall target — competitive at query time below roughly 50,000 vectors. The guarantee is completeness, not speed.
- WORM needs an external anchor against tail truncation. Publish the daily hash, or timestamp it, and the chain is complete end to end.
- BSL 1.1 is source-available, not OSI open-source. The code is auditable today; commercial rights convert to GPL v2.0 in 2036.