For your engineering team

What your engineers
actually have to do.

Written to be forwarded to a tech lead, not to impress a buyer. Where PlayerDesk runs, what it touches, what happens when a dependency fails, and who can switch it off. For the data path specifically, see the data & security brief.

See production patterns

Deployment footprint

PlayerDesk runs in your environment — your cloud account, your VPC, or your own hardware. It is not a multi-tenant service with your players sitting in somebody else's database, and there is no Arctura-side copy of your book to be breached, subpoenaed or sold.

Runs where you say

Containerised services deployed into your cloud account or on-premise. Your orchestration, your network policy, your secrets management.

Stateful data stays local

Conversation history, the token–identity mapping, audit and reasoning logs, the knowledge index — all in storage you own and back up.

Egress is one hop

Outbound traffic goes to one EU-region model endpoint, carrying redacted request text only. Everything else stays inside the perimeter.

No standing access for us

Arctura engineers have no permanent credentials to your environment. Access is granted by you, scoped, time-boxed and logged on your side — revocable without asking us.

What runs where

Five layers, two IP zones. The bottom three are yours and sit in your infrastructure. The top two are the part you would otherwise have to build.

"What you keep, what we bring, and where the boundary between them actually is."

Layer 5 — Dashboards & APIsOne screen per team. White-label optional.
Support team console Quality-control dashboard VIP team signal feed Compliance & audit view REST / GraphQL / Webhooks
Yours / white-label
Layer 4 — Data PlaneYour player data, your storage, your retention policy.
Player database (yours) Event stream (yours) Conversation history (yours) Redaction & token mapping Audit storage (yours)
Your infrastructure
Layer 3 — Adapter LayerPlugs into the PAM, helpdesk, payments and telephony you already run.
PAM (in production): IqSoft, EveryMatrix PAM (designed / on request): SoftSwiss, Pragmatic, Playtech, custom Helpdesk: LiveChat, Zendesk, Intercom, Freshdesk, custom Voice: Twilio, Vonage, SIP Payments: Stripe, Adyen, crypto KYC: Sumsub, Veriff, Onfido
Deployed in your env
Layer 2 — Domain LogiciGaming economics modelled in, not bolted on.
Bonus and wagering rules VIP qualification and pre-VIP signals Escalation policies RG escalation triggers Your QA protocol, encoded Configurable thresholds 20+ languages Workflow orchestration
Arctura core
Layer 1 — Core EngineThe AI you'd otherwise spend 6–12 months building.
LLM orchestration (SGR, multi-agent, tool use) ML model repository (VIP, risk, voice) Evaluation framework Audit & explainability Regulatory templates (UK / SE / NL / MGA)
Arctura core
Stays yours: your data, your PAM, your helpdesk, your stack. Ours to maintain: the AI core, so you don't build it.

One conversation, end to end

What actually happens between a player pressing send and an answer coming back — including the point at which a human takes over.

"Every routine chat handled by a human is cost you don't need. Every VIP handled by a bot is revenue you just risked. This is how one message gets resolved automatically — and how a VIP gets pulled out before that happens."

PlayerDesk conversation (multi-agent support)

Trigger: Player writes to chat — "I can't withdraw my money, what's wrong?" (any language, any platform).

  1. L3Helpdesk connector routes the message to PlayerDesk.
  2. L4Redaction pass tokenises identifiers before anything leaves the perimeter.
  3. L1Triage Agent detects intent (payout issue), language, urgency, VIP status check.
  4. L2VIP rules check → not VIP → continue with automated handling.
  5. L1Knowledge Agent retrieves relevant policies, common payout issues, KYC rules.
  6. L3Action Agent calls the PAM API → checks KYC status and pending withdrawals.
  7. L1Response Agent generates a contextual answer with the specific account state.
  8. L1QA Agent runs real-time evaluation (hallucination, tone, accuracy).
  9. L3Response sent back through the helpdesk connector.
  10. L4Audit engine logs the full conversation with its reasoning trace, in your storage.
  11. L1If complexity, RG language or a VIP threshold is hit → Escalation Agent prepares a summary and hands off to a human L2 or VIP manager.

Access, logging and incidents

  • No standing access. Arctura holds no permanent credentials to your environment. Support and incident access is granted by you, scoped to named individuals, time-boxed, and logged where you can read it — and revoked by you without involving us.
  • Everything is reconstructable. Each request records the original message, the redacted version actually transmitted, which fields were tokenised, the endpoint that answered, the response, and the reasoning behind the action taken. That trace is what your compliance team, your DPO and your regulator each need, and it is also what makes automated quality control possible.
  • Retention is your policy. Logs live in your storage under your rules. There is nothing on our side to delete, which is a simpler answer than a deletion certificate.
  • RG language is never automated. It is an unconditional escalation to a human, in every jurisdiction, regardless of model confidence.
  • Design-partner stage, stated plainly. At this cohort size you get direct founder-level engineering rather than a support portal — which means faster answers and no formal SLA document yet. We would rather say that than imply a 24/7 rota that doesn't exist.

The full data & security brief →

6–8 week deployment vs a 6–12 month in-house build.

Production-tested systems rather than research papers. Your data stays yours. Your stack stays in place. Bring your tech lead to the first call — the objections are more useful than the questions.