Model control in one place
KONEKT AI Gateway sits between your applications and AI models — local, custom, and public — according to company rules. It deploys on-prem or in a private cloud.
Applications and KONEKT Workspace connect to the Gateway over a secure VPN + TLS 1.3 encrypted tunnel. Routing, budgets, fine-tuning and security policies stay under your full control.
Gateway in your data center or rack.
Gateway in your private cloud environment.
3. Full provider abstraction & Custom Models
Swap the backend model without rewriting apps, or run purpose-built (fine-tuned) models with LoRA adapters trained on your company terminology.
One interface for applications
Applications connect to the Gateway. The backend model changes by policy, without rewriting business software.
Automatic routing by policy
The Gateway dynamically scores cost, latency, and the request's security level. Besides the base aliases, you create extra virtual models by department, confidentiality, budget, time, or fine-tuned model.
Extra virtual models under your conditions
The four base aliases (konekt-auto, konekt-fast, konekt-smart, konekt-private) are the starting point.
In the Gateway console you create more — the app still calls the same name, and rules change without code changes.
legal-smart, hr-private, it-coding
local only, masked, or a public model with permission
cheaper model after the limit, or a hard stop
local GPU if free, otherwise another allowed path
business hours, night batch, urgent requests
an alias to an internal model that passed evaluation
Full control of cost, access, and models
6. Privacy Policies
Automatic check whether content may leave to an external API or must stay on the internal network.
7. Department budgets
Monthly token and spend quotas for HR, marketing, sales, and legal — with alerts and hard stops.
8. Rate Limiting
Protection against congestion (RPM / TPM limits) and priority for critical services.
9. Detailed Usage Tracking
Real-time usage metrics: prompt tokens, completion tokens, latency, and cost per call.
10. KONEKT Console
A web admin panel for API keys, analytics, audit logs, and service configuration.
11. Private Models (GPU)
Integration with local vLLM models (DeepSeek, Llama, Mistral, Qwen, Kimi, MiniMax), treated as first-class endpoints in the catalog.
12. Strict Fallback Rules and Protection
Fallback rules keep operations running when a cloud provider is down (switch to an alternative model of the same class).
For konekt-private the rule is absolute: no cloud failover.
