askOdin — AI Judgment Infrastructure for Capital Allocation

SECURITY & DATA PROTOCOL

What We Hold, For How Long, and What Is Not Finished.

Private market diligence demands that a vendor be precise about what it holds and for how long. This page separates what we have implemented from what we have merely written down, because that is the distinction your risk officer is actually testing for. Your documents are never used to train a model. Raw documents are held no longer than 30 days: that ceiling is documented policy, automated enforcement of it is In Progress, and until it ships deletion is executed on request. There is no data-room connector or integration — files are uploaded for each audit.

// THE ARCHITECTURE

The Enterprise Security Architecture

Layer 01 — Bounded Ingestion

The Reality: Standard conversational tools utilize your unique inputs, deal metadata, and internal queries to train public neural networks. You are quietly leaking your fund's proprietary operational alpha to the public cloud with every prompt.

The askOdin Standard: The memos, financial models and data-room documents you upload are processed under Stateless API Orchestration and held no longer than 30 days under a documented retention ceiling. What survives that window is derived — the verdict, the judgment analysis, and structural data — never the documents themselves. That structural record can include identifying details, so it stays subject to erasure. As a matter of policy, we do not read your documents. That is a procedural control, not an architectural one: the system does not make staff access impossible, and we will not describe it as though it did.

Layer 02 — Tenancy, Stated Plainly

The Reality: Shared processing tiers expose sensitive transaction pipelines to cross-tenant leakage. This is a hard problem, and we will not pretend we have finished solving it when we have not.

The askOdin Standard: askOdin runs shared, and isolation of the processing cache is being hardened — that work is not complete. Single-tenant deployment, configurable processing region and deployment inside your own perimeter are on the roadmap and are not shipped. If any of those gate your firm, the honest answer today is that we are not ready for you, and we would rather tell you now than at the evidence request.

Layer 03 — Cryptographic Trail Generation

The Reality: A conversational chatbot history that outputs a slightly different response based on how a prompt is worded cannot survive a rigorous fiduciary audit.

The askOdin Standard: The verdict is computed outside the language model by a deterministic engine, so the compiled graph is invariant: bind the same variables and the same score comes back every time. Extraction is not frozen forever — it runs on an external model, and model versions move. That is why every audit is anchored to a SHA-256 of the exact files, and why a past result is verified by reading its record rather than by re-running it.

// THE SECURITY PROTOCOL

Built for the Risk Officer.

No Training Use

Your documents do not train anything — ours or anyone else's.

Extraction runs on the Google Gemini Developer API under a paid-tier agreement that prohibits training use, behind a swappable adapter. No customer document enters any training corpus, and askOdin does not fine-tune or distil a model on customer material. That is contractual and architectural, not aspirational.

Stateless API Orchestration

The language layer carries nothing between requests.

The language layer reads and extracts only. It never evaluates, and it carries nothing between requests. Every calculation and every verdict is produced by a statically-typed Go engine running against the extracted variables — the separation is what makes the audit trail mean anything. That is a statement about orchestration, not about tenancy: askOdin runs shared, and isolation of the processing cache is still being hardened (Layer 02 above).

Bounded Retention

Raw documents are held no longer than 30 days.

That ceiling is a documented policy. Automated enforcement of it is In Progress and ships with SOC 2 Type I — until then deletion is executed on request, and we would rather say so than call it a control. What persists past the window is derived: the verdict, the judgment analysis, and structural data. Never the documents. That structural record can include identifying details, so erasure rights reach it.

Cryptographic Anchoring

Every audit is bound to the exact files it read.

Submitted documents are fingerprinted with SHA-256 and the result is anchored to those hashes. That is what lets a third party confirm the deck they were sent is the deck that was scored, rather than a later revision wearing an older result. A fingerprint fails loudly when the file changes.

Edge & Transport

Inherited from Cloudflare, not asserted by us.

  • Edge & storage: Cloudflare WAF, DDoS protection, CDN, R2 object storage
  • At rest: AES-256, as provided by R2
  • In transit: TLS 1.3
  • API access: token-based auth with scoped permissions

Jurisdiction

Singapore — a recognized financial center under English Common Law.

  • Legal framework: Singapore Companies Act, PDPA compliance
  • IP protection: four U.S. provisional patents filed
  • Sub-processors: named publicly in Privacy Policy §5; supporting security documentation under NDA

Compliance Roadmap

A stated policy is not an implemented control, and an implemented control is not an evidenced one. Vendors who blur those three get caught at the evidence request. Here is which column each of ours sits in.

Implemented
  • AES-256 / TLS 1.3 encryption
  • Cloudflare WAF + DDoS protection
  • Documented 30-day retention ceiling
  • PDPA (Singapore) compliance
In Progress
  • SOC 2 Type I certification
  • Automated retention enforcement
  • GDPR Data Processing Agreement
  • Penetration testing (third-party)
Planned
  • SOC 2 Type II certification
  • ISO 27001
  • Single-tenant deployment
  • Configurable processing region
  • On-premise deployment option

// OBJECTION HANDLING

Security & Compliance FAQ

How our data is handled

Does askOdin train its models on the data we upload?

No. Extraction runs against an external model API under a paid-tier agreement that prohibits training use; the deterministic engine does the analysis. Raw documents are held no longer than 30 days under a documented retention ceiling. What persists beyond that is derived — the verdict, the judgment analysis, and structural data in the Judgment Graph benchmark corpus. That structural record can include identifying details such as founder names, so it remains subject to your erasure rights; systematic de-identification is on the roadmap, not shipped. No customer document enters any training corpus.

Where is our data processed — what data-residency options exist?

Today, one region: askOdin is operated from Singapore under PDPA, and document extraction calls an external model API (the Google Gemini Developer API, paid tier). A configurable processing region is on the roadmap and is not shipped. If residency in a particular jurisdiction is a gating requirement for your firm, tell us before you upload anything and we will tell you where that work stands.

Who are askOdin’s sub-processors?

Cloudflare for edge security and object storage, and the Google Gemini Developer API for document extraction, behind a swappable adapter layer. Extraction is the model's only role — it maps unstructured documents into typed variables, and the deterministic engine does the analysis. The sub-processors are named publicly in section 5 of our Privacy Policy; the security documentation behind them is provided under NDA.

How long are uploaded documents retained?

Raw documents are held no longer than 30 days. That ceiling is a documented policy; automated enforcement of it is In Progress and ships with SOC 2 Type I, and until then deletion is executed on request. What persists past the window is derived — the verdict, the judgment analysis and structural data — never the documents. That record can include identifying details, so your erasure rights reach it.

Does askOdin connect to our data room?

No. There is no data-room connector or integration. Files are uploaded for each audit, so what askOdin reads is exactly what you chose to send it, and the 30-day retention ceiling applies to every upload.

Contracts and certification

Is askOdin SOC 2 or ISO 27001 certified?

Not yet. Implemented controls: AES-256 at rest, TLS 1.3 in transit, stateless API orchestration, a documented 30-day retention ceiling on raw documents, and PDPA compliance. In progress: SOC 2 Type I, a GDPR Data Processing Agreement, and automated enforcement of the retention policy. On the roadmap: SOC 2 Type II and ISO 27001. We distinguish a stated policy from an implemented control and will tell you which is which — documentation is available under NDA for your security review.

Will askOdin sign a DPA, and do you use Standard Contractual Clauses?

A Data Processing Agreement with Standard Contractual Clauses for cross-border transfers is in preparation and is not yet available for signature. Contact the security team and we will tell you where it stands rather than quote a date we cannot evidence.

How does askOdin go through a corporate information-security review?

On the evidence, tier by tier. Every control on this page is labeled as stated policy, implemented or in progress, so your reviewer can test each claim at the level it is made. The sub-processors are named publicly in section 5 of our Privacy Policy; the security documentation behind this page is provided under NDA. Send your questionnaire or checklist with the request and we will answer it against the same tiers. SOC 2 Type I is in progress, not held, and the Data Processing Agreement is in preparation. If a certificate or a signed DPA is a gating requirement, raise it at the start and we will tell you where it stands. Request security documentation.

Deployment and audit trail

Can askOdin run as a stateless or dedicated instance?

Stateless, yes; dedicated, not yet. The runtime uses stateless API orchestration, and raw documents are held no longer than 30 days under a documented retention ceiling; the derived judgment record persists so the audit trail survives the document. Single-tenant deployment is on the roadmap and is not shipped. Tenant isolation on the shared processing cache is being hardened and is not yet complete — if strict isolation is a gating requirement for your firm, talk to us before you upload anything, and we will tell you exactly where that work stands.

Is the output auditable and defensible to a regulator?

Auditable and traceable, yes. Every verdict produces a Defensible Audit Log: each finding traced to the page or cell it came from, the Framework version that produced the score, and the SHA-256 fingerprint of the files read — a record you can put in front of an IC, an LP or an examiner. Whether that record satisfies a given regulator is the fund's determination, not ours; askOdin is not a regulated entity and does not certify compliance.

Questions about security?

Under NDA we provide the security documentation behind this page. Send your questionnaire or checklist with the request and we will answer it against the same tiers: stated policy, implemented, in progress. There is no SOC 2 report to send yet, and the DPA is in preparation; we will say so rather than send a placeholder.

For the technical evaluator: review the patent architecture.