The trust mark on AI consumption.
TokenOne is the AI wallet and delivery layer for productivity · and trust is built into every governed delivery decision, every burn, and every policy action.
Where we are. What we ship next.
Concise, honest, dated. We claim only what we have; we mark what’s in flight; we don’t front-run certifications we haven’t earned.
Last reviewed: 2026-05-12. Material changes are tracked in the changelog and announced on the news page.
At the network, not the perimeter.
Trust is only as strong as the controls baked into the rails. Every TokenOne call passes through the same governance pipeline regardless of which app made it or which provider delivered it.
Encryption everywhere
TLS 1.3 on the wire. AES-256 at rest. Provider keys (BYOK) encrypted at envelope, scoped per project, rotatable with a 24-hour overlap.
Bidirectional guardrails
Every prompt and every response screened · injection, PII, secrets, topic, leakage, toxicity, grounding. Standard on every plan.
Chain-verified ledger
Every call written to a signed BigInt ledger with full attribution back to the project. Replayable, tamper-evident, auditor-ready.
Policy as code
Compliance rules, residency, approved providers, approvals · versioned, scope-cascaded (global → tenant → project), replayable.
Region and residency
Pin AI delivery to UK / EU / US regions. Pin to specific approved providers. Residency gate runs pre-delivery.
Quality contract
Per-pattern validation thresholds drive delivery-path selection. Quality is cleared first; latency and burn are optimised second.
How TokenOne keeps governed delivery provable.
The control mechanisms behind every governed call · described in customer-safe, commercially relevant language. These are the components that make audit, replay, and savings reporting defensible.
Quality threshold logic
Per-pattern quality thresholds drive delivery-path selection. The most efficient path that clears the threshold wins; failures escalate to a stricter tier.
Prompt normalisation
Inbound prompts are matched, canonicalised, and rewritten · typically 35–55% token reduction · with a signed attribution trail back to the original input.
Signed ledger settlement
Every call writes to a BigInt ledger. Settlement is reconciled across providers. The ledger is the system of record, not a derived view.
Abstraction layer
Customers buy delivery by brand and tier, never raw provider SKUs. The abstraction insulates buyers from provider churn while keeping spend portable.
Replay and evidence trail
Any decision can be replayed against its policy version. Auditors get verifiable evidence of what ran, why, and what it saved.
Approval workflow
Policy violations escalate to an approval queue with named reviewers. Decisions are logged and tied back to the request that triggered them.
Mechanisms protected by UK patent filings · see footer for the full schedule.
Detailed pages, each kept canonical.
The hub pulls the headline; the detailed pages stay live for legal and security teams who need them under contract review.
Security details
Full security architecture, controls catalogue, threat model.
Service-level agreement
Uptime targets, response-time commitments, credits.
Live operational status
Component-level uptime, current incidents, scheduled maintenance.
Sub-processors
Every third party that processes data on our behalf, with scope and region.
BYOK details
Provider-key envelope encryption, scoping, rotation flows.
Reviewing TokenOne for your organisation?
We send DPAs, security questionnaires, sub-processor lists, and SOC 2 evidence on request. One contact, one packet.