Provenance for every character
Each span records its origin when it is created: a named person, a model and prompt version, a quoted evidence item, the tender itself, or “pasted, origin unknown”. Never guessed afterwards.
PPN 017 readyBuilt for the Procurement Act 2023
Every word in your bid.
Accounted for.
tood.os reads the whole tender pack, verifies every requirement against its source by code, and drafts answers where each sentence records who wrote it, which model, what evidence and who approved it.
No credit card required · Cancel any time · Google sign-in
01Tender packITT_Specification.pdf
Parsing text layer3 pages
02Requirementsverified by code
- pass/failCyber Essentials Plus at contract startp.1
- serviceNo loss of service in core hoursp.2
- formatQ4 max 750 wordsp.3
- pass/failDeclare whether AI tools were usedp.3
- blockedHidden instruction to AI ignoredp.3
28 found28 / 28 quotes matched
03Traced draftQ4 Transition
A named Transition Manager will lead a 12-week parallel run.“No loss of service during core hours.”We moved 1,400 users with no unplanned downtime.EV-031ISO/IEC 27001:2022 covers all transition work.EV-112
Approved by S. Okaforchain 9f2c…e41a
Invitation to tenderSpecificationProcurement Specific QuestionnairePricing scheduleSocial valueClarificationsContract termsFramework call-offDynamic marketCyber EssentialsService levelsWord limits
The problem
“Statements, facts or references that appear plausible, but are in fact false.”
How UK government guidance describes the risk of AI-written tenders. PPN 017, Cabinet Office
Allowed
Suppliers’ use of AI in bids is not prohibited.
Asked
Buyers may ask you to disclose where AI was used in your tender.
Checked
And to confirm the output was checked and verified for accuracy.
Most AI bid tools sell speed. tood.os sells the audit trail. Speed is a side effect.
Live demo · real model, real verification
Run it on a tender. Right now.
Pick the sample or drop your own PDF. A model extracts every requirement with a verbatim quote. Then plain code checks each quote against the document. Anything it cannot find is rejected, not trusted.
Results appear here. Each requirement links to the exact sentence it came from, highlighted in the document.
Provenance view
Click any sentence. See exactly where it came from.
Person, model, evidence or tender. Unsupported claims block export until someone removes them or attaches proof.
Q4Describe how you will maintain continuity of service during transition.
We will run a 12-week parallel transition, led by a named Transition Manager who reports weekly to the Authority’s contract lead. This directly addresses the Authority’s requirement for “no loss of service during core hours (08:00–18:00)”. In our 2024 migration for a regional housing association, we moved 1,400 users to a new service desk with no unplanned downtime.EV-031 Our ISO/IEC 27001:2022 certification covers all transition activities.EV-112 We guarantee 99.99% availability throughout the contract.
Written by a personDrafted by AI, evidence-boundQuoted from your evidenceQuoted from the tenderUnsupported claim
1 unsupported claim. Export is blocked until it is resolved.
The tood.os standard
Nine rules the product enforces. Not a slogan.
They apply whoever writes the text, a person or any AI model, and they apply to our own engineering as well.
Requirements tied to the source
Every requirement carries a verbatim quote, located in the original by code. You just watched it happen.
Claims tied to evidence
Figures, clients, certifications and past results must link to your evidence library or they are flagged.
A person is always accountable
A model can draft. It can never approve. Every answer needs a named approver.
Tamper-evident history
Append-only, hash-chained revisions. Rewriting history is detectable.
Disclosure-ready
The AI use record is generated from the ledger, never written by a model.
Models earn their place
No model is used until it passes admission tests, re-run on every version change.
Stays within bounds
Only your tender and your evidence. No pricing commitments, no legal advice. When it does not know, it says so.
We are auditable too
A public trust centre will list the models we use, their latest test results, any fabrication incident that reached a customer, and every sub-processor that touches your data.
AI use & verification record
When the buyer asks, you already have the answer.
Contracting authorities may ask whether AI helped write your tender and whether it was checked. tood.os answers from the revision ledger, question by question, assembled by code from what actually happened.
- Which questions used AI, and which did not
- How much AI-originated text survived human editing
- Who verified each answer, and when
- A chain-head hash tying the record to the exported file
Sample record · fictional
AI Use & Verification Record
Northgate Digital Ltd
Service Desk & EUS
Service Desk & EUS
| Question | AI used | AI text kept | Verified by | Date |
|---|---|---|---|---|
| Q1 Understanding | No | 0% | J. Patel | 15 Sep |
| Q2 Methodology | Yes | 38% | S. Okafor | 16 Sep |
| Q3 Team & CVs | Yes | 12% | J. Patel | 16 Sep |
| Q4 Transition | Yes | 27% | S. Okafor | 16 Sep |
| Q5 Social value | No | 0% | M. Reyes | 17 Sep |
rev 40a71d…03be
rev 41c0e8…9a12
rev 425b3f…d7c0
rev 439f2c…e41a
Model admission
No model works here until it passes.
Results are kept as a model card per version. Production pins the admitted version; a provider update does not reach your bids until it is re-tested. These are launch targets; measured results will be published.
Fabrication baitAsked to cite certificates that are not in the libraryrefuse or flag 100%
Prompt injectionHidden instructions inside tender documentsignore 100%
Honest gapsQuestions the evidence cannot answersays so ≥ 95%
Requirement recallExpert-annotated public tender packs≥ 95%, pass/fail ≥ 98%
ScopePricing, legal advice, outside factsrefuse or label ≥ 98%
ConsistencySame input, five runs≥ 95% overlap
What we do not promise
Honesty about the limits is part of the standard.
- We do not guarantee your evidence is true.We show a claim is traced and supported. If a case study is wrong, the approver is accountable.
- We do not guarantee you will win.We help you submit something complete, compliant and defensible.
- We do not use AI detectors.They guess. We record origin when text is created, and label pasted text as unknown.
- We do not submit on your behalf.A person in your company submits. Always.
Pricing
We do not compete on price. We price for being right.
Founding pricing for the first 30 companies, locked for 12 months, in exchange for honest feedback.
Starter
£99 / month, excl. VAT
One tender pack, full requirement matrix with verified source quotes.
- 3 active tenders
- 2 user seats
- Requirement matrix
- Audit trail
Most popular
30 founding places
Professional
£299 / month, excl. VAT
The full workspace for one company, with provenance and the AI use record on every bid.
- 20 active tenders
- 10 user seats
- Gemini AI drafting
- Provenance view and export gate
- AI use & verification record
- Price locked for 12 months
Enterprise
£799 / month, excl. VAT
For large organisations with unlimited tenders, SSO, and custom data residency.
- Unlimited tenders & users
- SSO / SAML
- Custom data residency
- Dedicated SLA
Beta pricing. Plans may change before general availability; founding members keep their price for 12 months.
How the 14‑day trial works
Experience the full platform, zero risk.
Try every feature with your own tender documents. No payment details required.
1
Sign up in 30 seconds
Use your Google account. No credit card required. Start immediately.
2
Upload your first tender
Drop any PDF tender pack. The system parses requirements in minutes.
3
Explore the workspace
Try AI drafting, audit trails, verification, and export.
4
Decide if it fits
Use the full platform for 14 days. Cancel anytime with one click.
Questions
Asked plainly. Answered plainly.
Is it allowed to use AI to write a public-sector bid?
Yes. Cabinet Office guidance states that suppliers’ use of AI is not prohibited, but authorities may ask you to disclose it and to confirm the output was checked for accuracy. See PPN 017.
Where is my data processed?
The product is designed to run in the UK (AWS London), with AI inference routed only to providers and regions that meet our residency rules. Your documents are never used to train models. The public demo on this page processes files in memory and does not store them.
Which AI models do you use?
Only models that have passed our admission tests. We are provider-neutral; the admitted list and model cards will be published in our trust centre.
What happens to text I paste from elsewhere?
It is labelled "pasted, origin unknown" and counted separately in the AI use record. We do not guess where it came from.
What do founding members commit to?
Using tood.os on at least one real bid, and a short feedback call each month. In return you keep founding pricing for 12 months and shape the roadmap.
Founding cohort
Bid with AI. Stand behind every word.
We are onboarding 30 UK companies that bid for public contracts. Tell us a little about you and we will be in touch personally.