# Intramedial: short security overview (draft)

Version of 1 November 2026.
Current copy: https://intramedial.com/trust/security-pack

## About this document

This is a short draft, not a full whitepaper. It states what the repository documents today and says where it stops. It claims no certification, audit, penetration test or service level, and it is not a contract: the legal documents are the binding text.

## What the product does

Intramedial is an agent coordination and admission plane. An agent declares an intent: an actor, a resource and a verb. Intralock, a pure function with no network call and no model, admits the intent under a lease or refuses it with a reason. Decisions are appended to a ledger under Merkle roots. A range of the audit chain can be sealed into a bundle signed with Ed25519, which can be checked offline.

## Approval and control

* A repository write that needs confirmation is refused with confirmation_required until an approver has issued a token (github.enforce).
* Eight scopes need two different administrators. The gated route runs only with a warrant for that scope and body, and breaking the seal is written to the ledger first (dual-control.scopes).
* Agents hold no standing cloud keys. iam.assume issues short-lived credentials against a live lease, and their lifetime never outlasts the lease.
* Halt and quarantine stop or isolate agents.

## Regions and residency

* Seven regions are deployed: US East, US West, Europe Central, Europe West, Europe North, Asia-Pacific and Oceania. The three European regions store under Cloudflare's EU jurisdiction; the others use a location hint. The workers and storage that hold customer content are mirrored in each region.
* The Europe region is created under Cloudflare's EU jurisdiction. Asia-Pacific and Oceania use a Cloudflare location hint, which is a request and not a guarantee.
* Compute runs at the nearest Cloudflare location to the caller. Regional Services has not been bought. KV, Queues and Workers AI cannot be pinned; they hold no customer content.
* Billing runs in US West. The Control Plane and Provider Admin are single and not regional.
* Rehoming copies a snapshot to another region, which applies it on its next request. Nothing is deleted at the source then, the snapshot is not signed and there is no restore action.

## Bundle signing

* A bundle names its key id and jurisdiction. The private keys are Worker secrets set per region. They are not held in an HSM or AWS KMS.
* The signer prefers the region's key and may fall back to another key it holds. With no key at all, the snapshot answers 503.
* The public keys are documented at /.well-known/ledger-bundles-keys.json. Live serving has not been confirmed.
* No rotation schedule, key expiry, revocation list or transparency log is documented.
* The offline verifier is in the repository (verifier/), MIT licensed, and is not on a registry yet.

## Vulnerability disclosure

Reports go to security@intramedial.com or the form at /security/report. A security.txt file is published at /.well-known/security.txt. No monetary bounty is offered.

Targets, under the responsible disclosure policy: acknowledgement within 3 business days, initial assessment within 10 business days, updates at least every 14 days, a fix for critical and high severity aimed at 30 days and for lower severity at 90 days. They are targets, not guarantees.

Personal data breaches: the account owner is notified by email without undue delay and in any case within 72 hours of Intramedial becoming aware of a breach affecting customer personal data, under the Data Processing Addendum. Written answers to a reasonable security questionnaire are given within 30 days under DPA section 9 (privacy@intramedial.com).

## Recovery objectives

Exit objectives publish a recovery time and a recovery point for five data classes. None has been measured: every class reads unverified until a drill is recorded.

## What is not yet published

Credential storage and lifetimes, session settings, rate limits, tenant separation inside a region, encryption at rest, retention periods, deletion and erasure handling, availability and a status page, service levels, sub-processor terms, personnel controls, penetration tests and certifications. No statement about any of these is made here.

## Sources

| Statement | Public page | Source in the repository |
| --- | --- | --- |
| Admission and the ledger | /docs/coordination-admit, /docs/ledger-blame | content/tools-docs/coordination-admit.mdx |
| Bundle signing and keys | /docs/ledger-bundles-snapshot | content/tools-docs/ledger-bundles-snapshot.mdx, runtime/ledger-bundles/src/keyring.ts |
| Approval and dual control | /docs/github-enforce, /docs/dual-control-scopes | content/tools-docs/github-enforce.mdx, dual-control-scopes.mdx |
| Short-lived credentials | /docs/iam-assume | content/tools-docs/iam-assume.mdx |
| Regions and rehoming | /regions, /docs/rehoming-plan | docs/regions/REGION-MODEL.md, content/tools-docs/rehoming-plan.mdx |
| Recovery objectives | /docs/exit-objectives | content/tools-docs/exit-objectives.mdx |
| Disclosure contacts | /security/report | public/.well-known/security.txt |