Platform

The whole stack,
already assembled.

The components and the features included, on your Kubernetes cluster.

Portal, gateway, billing and data run on your Kubernetes. Models are cloud or local, behind the same policy and accounting boundary. Manage everything from the portal or with venticli and declarative manifests.

Always includedOptional — only if you need it

People and applications

Browser, API and terminal

IT/EN portal

Teams, quotas, credit, keys, model catalogue, remote agents, audit and API lab.

Compatible APIs

Chat Completions, Responses, Messages, FIM, token counting, embeddings, vision, speech and transcription, with unbuffered SSE.

venticli and OpenCodeOptional — only if you need it

A CLI for declarative resources with browser or device sign-in; an OpenCode launcher with a dynamic catalogue of authorised models.

Open WebUI chatOptional — only if you need it

Encrypted conversations, MCP and private search, bound to the tenant; it keeps its own branding and licence.

Portal and API on the same gateway

Gateway and governance

App and ops namespaces

Control

Identity, tenants, catalogue, policies, API keys, remote agents and runtime jobs from a fixed catalogue.

Gateway and Switchyard

Native protocols, routing and choice of the most suitable model among authorised candidates, ranked by exact Decimal cost.

Billing

Wallets, reservations, a balanced ledger, top-ups by Stripe or bank transfer, invoices and reseller commissions.

PIIOptional — only if you need it

Local detection of personal data in text, reversible placeholders and encrypted maps.

Policy, credit and PII before the model

Data and security

Data and edge namespaces

PostgreSQL and Valkey

CloudNativePG with Barman backups in the bundled profile, or external services; Valkey for rate limits and ephemeral state.

WAF and TLS

Traefik, Nginx with ModSecurity CRS, cert-manager certificates; default-deny NetworkPolicies.

Keycloak and OpenBao

OIDC sign-in with Google, passkeys and TOTP; secrets and infrastructure credentials held in OpenBao.

Outbound agent — no open port

Model runtimes

On your hardware or on VMs Ventic creates

Remote agent

A Rust agent for Linux and Windows: outbound WSS connection, pairing with a 15-minute invitation, administrator approval.

Certified runtimes

Strata and llama.cpp are tested; vLLM, Ollama and SGLang are accepted but not yet certified.

Managed inferenceOptional — only if you need it

Compute pools and deployments through OpenTofu with approved plans, routing and authorised autoscaling. One VM, one complete replica.

Licence verified, accounting always available

Providers and licence

External services

DeepSeek and OpenRouter

Cloud models with public names, permissions and prices managed in the catalogue; verified tariffs and FX kept apart from public prices.

Licman

An Ed25519-signed lease of at most 72 hours. On expiry new requests stop, but settlement, reconciliation, export and erasure remain available.

SMTP, S3, StripeOptional — only if you need it

Sign-in and invitation email, data archive and backups, online payments: provider integrations.

The chart does not install cluster operators (KEDA, cert-manager, Traefik, CloudNativePG): the cluster administrator provides them. An installation uses four namespaces and does not consume an extra licence per replica or Job.

What you get

Everything you need
to govern AI.

Book a technical call

Prepaid credit

A wallet per company: reserve before every request, settle on measured usage. Credit arrives only once payment is confirmed.

Group quotas

Allowances for people and groups, recurring allocations, reseller commissions calculated on settled usage.

Governed API keys

Allowed models, per-request maximum, budget, expiry and revocation. Current policies also apply to clients holding an old catalogue.

Reversible PII

Placeholders before the model, originals restored in the stream. Optional, text only: images and audio need specific policies.

Switchyard

Chooses among authorised models by suitability, effort and exact Decimal cost. Selection calls are accounted for.

Remote agents

Runtimes on your hardware over outbound WSS, with approval and explicit consent before sharing with other companies.

Managed inference

Compute pools and deployments with OpenTofu, leastBusy, weighted or failover routing, and autoscaling within approved limits.

Multi-tenant and reseller

Customer companies as application tenants, reseller delegation and commission agreements, without a cluster per customer.

Chat and integrations

Open WebUI with encrypted conversations, OpenCode with a dynamic catalogue, venticli for declarative resources.

Export and erasure

Encrypted content, JSON export, erasure requests and an optional archive of future inference to the company's S3.

Observability and audit

OpenTelemetry metrics, logs and traces, Perses dashboards filtered by company and an append-only activity log.

Licence and continuity

A signed Licman lease of at most 72 hours. On expiry inference stops; settlement, export and erasure remain.

Performance and security

Written in Rust.
Fast, scalable, protected.

Performance, mission-critical reliability and cybersecurity are not add-ons for later: they ship with the install. The code is fast and safe by construction, scales horizontally, and traffic goes through a web application firewall that is already configured.

The core of the platform is Rust

Gateway, control plane, billing, worker, the agent on GPU servers and venticli are written in Rust, the language chosen for the critical components of operating systems, browsers and clouds.

Memory safe: No buffer overflows or use-after-free: the most exploited vulnerability classes are ruled out by the compiler.
Fast and lean: No garbage collector: low, steady latency on streaming, modest CPU and memory use.
Signed binaries: Releases verified by signature, agents with installation-bound identity.

Built-in WAF

Every request to portals and APIs goes through Nginx with ModSecurity and the OWASP Core Rule Set, configurable from the admin console.

OWASP Core Rule Set v4Blocks SQL injection, XSS, remote code execution, path traversal and protocol violations.
Tunable anomaly scoringConfigurable paranoia level and anomaly threshold; detection-only or blocking mode.
Per-IP rate limitingSeparate limits for login, API and portals, with burst and requests per minute: stops brute force and abuse.
Connection limitsConcurrent requests per IP and in total, maximum body size: protects against spikes and application-level DoS.
Blocklists and threat feedsIPs and networks blocked by hand, plus public lists of malicious IPs, with exceptions for your networks.
Defence in depthAutomatic TLS, default-deny NetworkPolicies and GPU agents with no inbound ports.

High performance, built for mission-critical use

Ventic is designed for mission-critical workloads: it carries sustained traffic from whole teams of developers, agents and applications calling models all day long, stays up when a node or provider fails, and grows by adding replicas instead of buying bigger servers.

Asynchronous I/ORust gateway on Tokio: many concurrent streams per replica, without a thread per request.
Horizontal scalingGateway, control and billing are stateless: state lives in PostgreSQL and Valkey, so you add replicas behind the load balancer. Two replicas by default.
GPU autoscalingCompute pools grow and shrink on their own with load, on your hardware or in a neocloud.
No single point of failureSeveral replicas per service and several targets per model: if a node or provider fails, traffic moves to the others.

Get started

Get in touch
to get started.

We help you install Ventic on your infrastructure and start directly from the workload you want to put in production.