Security & Trust
Security & trust, by design
Crelis is built to be handed to a security review. It runs as a hosted service, decides on decision metadata rather than raw customer data, and records every decision in a tamper-evident audit trail.
Deterministic decisions. Short-lived execution visas. Tamper-evident evidence.
- Deployment
- Hosted service (SaaS)Live
- Posture
- deployments start in shadow posture and move to enforcement by agreement
- Data
- Decision metadata, not raw data
- Evidence
- Tamper-evident audit
- Isolation
- Server-enforced isolation with a production guard
- IP
- Patent-pending · Runtime authorization · 2026
Trust boundary
Beside your systems, not inside your model
Crelis sits beside your systems, not inside your model. It runs as a hosted service; deployment options beyond it are on the roadmap.
Trust boundary
Crelis sits beside your systems, not inside your model
Your systems
AI agent
proposes a consequential action
Calling system
sends decision metadata — categories and outcomes, not sensitive values
Authorized reviewer
signs off when policy requires human approval
Crelis · hosted service
Deterministic policy evaluation
allow · require human approval · escalate · block
Tamper-evident audit record
show what was decided, and prove it has not been altered since
Server-enforced tenant isolation · keys held server-side · TLS 1.2+ in transit · provider-managed encryption at rest
Your endpoint
Endpoint verifies visa
signature valid · short-lived · single-use
Your system executes
the visa and the evidence make the approval and the execution the same record
Stays with you
Raw identity, card, or document numbers · customer names, contact details, or message contents
Keys
Keys are held server-side and never exposed to the browser
Integration credentials
Sealed at rest and never shown back in plain text
Tenant isolation
Server-enforced isolation with a production guard. Cross-tenant access is denied — not merely hidden in the UI.
Decision authority
Deterministic policy evaluation decides; no AI model makes the final call.
Thresholds are yours to set; the final decision is never made by an AI model.
Data handling
Decision metadata, not raw customer data
Data class
Stored by Crelis
Decision metadata
action types, results, policy outcomes
Trust signals as types/results
e.g. “identity: not verified”
Tamper-evident audit records of each decision
Raw identity, card, or document numbers
Customer names, contact details, or message contents
Any field the calling system doesn’t send for a decision
In transit
all traffic uses TLS 1.2+
At rest
provider-managed encryption applies
Keys
held server-side and never exposed to the browser
The shape is structural — a Crelis decision is made on categories and outcomes, not sensitive values, so the integration keeps raw values out of the request rather than relying on a setting.
Evidence integrity
Each consequential action leaves evidence
Who proposed it, which inputs were evaluated, which policy fired, who approved it, what executed — every decision reproducible from its recorded context and exportable for auditors and regulators.
- inputs
- policy version
- reviewer sign-off
- visa
hashed into a tamper-evident audit trail you can export
Every decision is written to a durable, tamper-evident record — show what was decided, and prove it has not been altered since.
Reproducible decisions — Same evidence. Same outcome.
How this maps to Singapore's MAS SAFR white paper: our breakdown of the records SAFR implies
Audit event stream · illustrative
Illustrative datasealed
14:02:11Z · Finance Agent
Proposed wire transfer $250,000 → ACME-7741
14:02:11Z · Crelis
Inputs evaluated · policy WIRE-HIGH fired · REQUIRE HUMAN APPROVAL · risk context 91
14:02:12Z · Crelis
Routed to treasury reviewer pool (2 eligible) · visa withheld
14:06:48Z · Treasury reviewer
Checked counterparty, approved with note
14:06:49Z · Crelis
Execution visa issued · signed, single-use
14:06:49Z · Banking endpoint
Visa verified · wire released — ref TXN-88231
14:06:49Z · Crelis
Evidence recorded · hash 0x91f3…aa07
- Tamper-evident by construction
- chained into a tamper-evident record
- Outage
- consequential actions fail closed — if authority can’t be determined, they wait and are never silently allowed
- Human oversight
- an unrecognised action escalates to a human instead of executing
- Execution visa
- Bound to one action, one resource, one identity
Shared responsibility
What Crelis does, and what stays with your team
Responsibility
Crelis
Your team
Set policy thresholds
“Thresholds are yours to set; the final decision is never made by an AI model.”
Your team
Decide the outcome
“Crelis independently decides whether a consequential action is allowed, needs approval, or must be blocked.”
Crelis
Keep raw values out of the request
“A decision needs the category and the outcome, not the sensitive value, so your integration keeps raw customer data out of the request.”
Your team
Approve when policy requires
“When policy calls for it, the action waits for an authorized reviewer, whose sign-off becomes part of the record.”
CrelisYour team
Issue or withhold the execution visa
“Crelis evaluates it against policy and decides — then issues or withholds the execution visa and records the evidence.”
Crelis
Verify the visa and execute
“Visa verified by your endpoint. Your system executes.”
Your team
Seal and export the evidence
“Each decision — inputs, policy version, reviewer sign-off, visa — is hashed into a tamper-evident audit trail you can export.”
CrelisYour team
Tenant isolation and hosting
“Runs in our environment with server-enforced tenant isolation. In transit, all traffic uses TLS 1.2+; at rest, provider-managed encryption applies.”
Crelis
Deployment
Where it runs today, and where it is going
Crelis sits beside your systems, not inside your model. It runs as a hosted service; deployment options beyond it are on the roadmap.
Model
What it means
Status
Hosted service (SaaS)
Runs in our environment with server-enforced tenant isolation. This is where customer tenants run.
Deployment options beyond our hosted service
Dedicated and customer-controlled deployment models are on the roadmap — not available today. Crelis is delivered as a hosted service only.
Crelis runs as a hosted service. Deployments start in shadow posture and move to enforcement by agreement.
Disclosure
What is claimed, and what is not
Claimed
- Runtime authorization for AI agents. Crelis independently decides whether a consequential action is allowed, needs approval, or must be blocked.
- Deterministic decisions. Short-lived execution visas. Tamper-evident evidence.
- Decision metadata, not raw data.
- Server-enforced isolation with a production guard.
- Patent-pending · Runtime authorization · 2026.
Not claimed today
- Dedicated and customer-controlled deployment models are on the roadmap — not available today. Crelis is delivered as a hosted service only.
- Results from the Crelis test environment; not customer-production data.
- Every visual on this site is labelled illustrative data.
Reviewing Crelis for your team?
Request the full Security & Trust Pack, or send your security questionnaire — we’ll turn it around quickly.
Found a vulnerability? Responsible Disclosure
