Trust center

What Secure Raven actually does.

Connection modes, encryption, retention, subprocessors and release gates, without claiming a HIPAA certification, a finished SOC 2 report or a completed pen test.

Read the trust manifest Security at Secure Raven

Manifest: Machine-readable copy of this page: /api/trust/manifest. Version 2026-10-17.1.
Claims: 6 fields, every one false.
BAA: Connecting a source does not create one.
Release gates: 0 of 5 open in production.
Pen test: Not commissioned. The scope is written.

Disclosures

Claims we do not make.

What we have not done yet, and what we choose not to do. Written down, so you can hold us to them.

Not yet
We are not SOC 2 audited yet.
Our Type I target scope is Security and Confidentiality. No auditor is engaged yet.
No such thing
No one is “HIPAA certified”, including us.
HIPAA Hawk maps technical safeguards. It does not certify.
Not yet
No penetration test has been completed yet.
The trust manifest says so too.
By design
We are not a WAF, and we do not proxy your traffic.
Shield is middleware and SQL you install and approve in your own app and database.
By design
We do not use AI to find or rank issues.
Fixed rules produce every finding.
By design
We do not store your rows.
We keep table names, policy names and flags.

Trust manifest

live · machine-readable

GET /api/trust/manifest
"claims": {
  "hipaaCertified": false,
  "baaOffered": false,
  "connectingCreatesBaa": false,
  "soc2IndependentlyAudited": false,
  "externalPentestCompleted": false,
  "sitsInFrontOfSubscriberApi": false
}

Every certification field in the manifest is false. A test checks that, and so does the server on every request.

Read the trust manifest

Connection modes

What we hold, mode by mode.

How each kind of connection reaches your systems, and the credential Secure Raven keeps for it, if any.

Connection modes

5 modes

Connection modes, what Secure Raven holds for each, and what each is the default for
Mode	Held by Secure Raven	Default for
Public website scan`public_web`	None	web
Subscriber-hosted push probe`push_probe`	Ingest token hash and an Ed25519 public key only. Secure Raven generates the key pair when the token is created, shows the private key once and does not store it. No database credential.	supabase recommended
Restricted Postgres scanner role`restricted_role`	None: restricted-role login is not offered while Gate B is closed	not offered
Privileged Supabase API mode`privileged_api`	service_role or sb_secret_ key	optional, highest risk, off in production
Shield apply`shield_apply`	None by default. Subscriber pastes digest-bound SQL.	never implicit

Encryption

stored credentials

Envelope: AES-256-GCM envelope v3 bound to tenant, source, kind, and format.
Production key: None at launch: storing database credentials stays off until an external KMS holds the key. A local KEK is refused in production unless ENVELOPE_ALLOW_LOCAL_KEK is explicitly set, and never when ENVELOPE_REQUIRE_KMS is set.

What we retain

Kept for a set time, then purged.

The same schedule for every tenant and every plan. The audit log is the exception: we do not delete it. The current chain began on 2026-10-16, when we rebuilt our database, and nothing in it links to the earlier chain.

Retention

days, on one axis

Events90 days, then purged
Raven runs, once finished365 days, then purged
Resolved findings730 days, after resolve or accept, then purged
Audit logNot deleted, chain since 2026-10-16

Read from the trust manifest. A tenant audit export carries the chain’s head hash and counts only.

Subprocessors

Who else touches the data.

The services that process Secure Raven data for us. Anything not listed here has no data-processing relationship with us.

Subprocessors, what each does for Secure Raven and what data it handles
Subprocessor	Purpose	What it handles
Supabase	Postgres, Auth and Edge Functions, in US East (us-east-1)	Your account and tenant data: systems, findings, events and the audit log. Requests to Supabase pass through Cloudflare, Supabase’s own provider.
Vercel	Hosting for this site and the dashboard, with functions in US East	Requests and responses in transit, and server logs.
Resend	Transactional email: invitations, password resets, account security and incident alerts	Recipient address and workspace name, one-time sign-in links and codes, and for incident alerts the severity, a short title and a dashboard link. A title is a fixed phrase and may include the start of a finding’s title. Nothing else from findings.
Olyron	Sign in with Olyron, if you choose it	Your identity and email address.
GitHub	Source code and CI	Our source code. No customer data.

No AI or LLM vendor receives findings, events, credentials or the audit log.

Release gates

0 of 5 gates open in production.

Code being finished is not the same as a gate being open. Each gate opens only after its review, test or sign-off.

Gate A: Safe beta
Code completeProduction off
Public web scan and subscriber-hosted push probe. Privileged credentials stay off in production by default.
Gate B: Restricted direct scan
Code completeProduction off
Restricted scanner role and MFA/step-up exist in code. Independent SQL review is still required before calling this gate open.
Gate C: Privileged Supabase API mode
BlockedProduction off
Blocked until an external pen test closes critical/high findings. Privileged mode is optional and off in production unless explicitly enabled.
Gate D: Remote Shield apply
BlockedProduction off
Default apply is subscriber-executed SQL bound to the reviewed pack digest. REMOTE_SHIELD_APPLY_ENABLED stays off.
Gate E: Regulated data
BlockedProduction off
REGULATED_DATA_CONNECTIONS_ENABLED stays off until BAA, subprocessors, and counsel approve HIPAA operational scope. A signed BAA is not a certification.

Production defaults

fail closed

Privileged database connections: Off
Regulated-data connections: Off
Remote Shield apply: Off
Signed ingest required: On
Step-up before privileged actions: On
Shield apply on scan credentials: Off

Service levels

SLOs we will measure.

Targets only. None of these is a live product metric yet.

Service level objectives: what each measures, its target, and whether it is measured yet
SLO	What it measures	Target	Measured
`credential.unwrap_success`	Authorized broker unwraps of live, non-erased envelopes	99.9% excluding revoked or disconnected sources	Not yet
`credential.time_to_revoke`	Disconnect to claim/unwrap refusal	Next claim cycle; queued jobs skipped	Not yet
`scan.isolation`	Scan grants never include shield.apply	100% of planned and claimed runs	Not yet
`ingest.signature_rejection`	Unsigned or tampered ingest is rejected (Ed25519 signature required on every request)	100% of fail-closed production traffic	Not yet
`shield.rollback_time`	Time to stop remote apply and keep subscriber SQL as the path	Flag off immediately; no global remote apply	Not yet

Disclosure

Report a vulnerability.

Email security@secureraven.app. Do not include health information, credentials or customer data.

Read the disclosure policy security.txt

Bring the Ravens to your own systems.

Secure Raven is in private early access, by invitation. Tell us what you run.

Request early access Compliance

What Secure Raven actually does.

We are not SOC 2 audited yet.

No one is “HIPAA certified”, including us.

No penetration test has been completed yet.

We are not a WAF, and we do not proxy your traffic.

We do not use AI to find or rank issues.

We do not store your rows.

Trust manifest

Connection modes

Encryption

Retention

Gate A: Safe beta

Gate B: Restricted direct scan

Gate C: Privileged Supabase API mode

Gate D: Remote Shield apply

Gate E: Regulated data

Production defaults

Report a vulnerability.

Bring the Ravens to your own systems.