Architecture
Fullstack architecture
Every component runs inside the customer-controlled environment. Each layer is independently hardened, monitored and audited.
Customer Local Zone
All components above run inside the customer-controlled perimeter. SSO/RBAC, HMAC webhooks, audit trail and observability are included at the node level.
AI Runtime
AI provider options
The AI Gateway routes analysis through the provider configured for each environment — no external calls without explicit customer choice.
Mock Provider
For deterministic demos, tests and offline validation.
Frontier API
For non-sensitive hosted demos and evaluation environments.
vLLM Local
For customer-controlled open model inference.
LouhiGPT Local
Designed to support LouhiGPT Local once endpoint, model and runtime settings are configured through the AI Gateway.
LouhiGPT-ready local AI architecture. TALKTO+ Enterprise AI Gateway is designed to support LouhiGPT Local, vLLM-compatible models or other customer-approved local LLM runtimes. LouhiGPT-specific deployment is configured through endpoint, model and runtime settings once customer and model details are available.
Compute
Customer GPU ready
TALKTO+ Enterprise is designed to run with customer-controlled GPU capacity, including local AI inference profiles such as vLLM-compatible runtimes and NVIDIA-accelerated environments such as DGX Spark.
GPU acceleration and hardware configuration are the responsibility of the customer environment. TALKTO+ provides runtime-agnostic API integration.
Deployment modes
TALKTO+ Core
Managed SaaS baseline for pilot and evaluation.
Single-node Enterprise Local Node
Full stack on a single customer-controlled VM or server.
Customer VPC / VM
Deployed inside customer cloud or on-premise virtual machine.
Kubernetes / OpenShift
Container-native deployment for regulated enterprise environments.
Intelligence
Internal Enterprise Swarm
TALKTO+ aggregates only human-validated, anonymized patterns inside the enterprise boundary — creating an internal learning layer without any external raw data sharing.
Uses validated findings, not raw conversations
No raw strategy text leaves the local node
No gradients are shared externally
No model weights are shared externally
No cross-company sharing by default
Global opt-in requires separate agreement and feature flag
Global Opt-In Hub — disabled by default. Requires explicit customer agreement and feature flag activation. Shares only anonymized aggregated patterns. No raw text, no model gradients, no model weights are included.
Engagement
Enterprise pilot path
Diagnostic Workshop
2–4 hour executive risk framing and signal map.
Activation Pilot
Controlled pilot against one transformation program.
Enterprise Local Node
Zero-egress deployment with customer-approved AI runtime.
Boardroom Operating Model
Recurring validated risk reviews, action cards and reporting cadence.