Demo
€030 days
Install the full stack on your server and try it on your real workloads. Then move to the licence without reinstalling.
Deploy the demoPricing and comparisonSave, and
Save, and
govern AI.
Ventic is a product: you pay a monthly licence, per host or per company, on your own infrastructure. The licence doesn't depend on tokens; model consumption is prepaid credit you govern per group and per key.
Licence price listThe price list,
The price list,
no surprises.
Ventic is bought as a monthly software licence, per host or per company, at the same prices as Licman. Prices in EUR, VAT excluded, one-month commitment.
Per host
€29/ month + VAT
One licence per host running Ventic. Ideal to start with one or two servers.
Buy on LicmanPer companyUnlimited hosts
€299/ month + VAT
All of the company's hosts under one licence. Better value as the fleet grows.
Buy on LicmanInitial prices in EUR, VAT excluded, as of 28 October 2026. Changes apply to new requests; quotes and subscriptions already agreed keep their amount. Free 30-day demo. Looking for a managed service? It is sold separately on llm.netter.io.
Comparison
Ventic vs the alternatives
YesPartlyNo
| Criterion | Frontier subscription | Pay per token | Ventic On Premise |
|---|---|---|---|
| Predictable end-of-month cost | Flat fee | Depends on usage | Fixed licence; AI usage is separate prepaid credit |
| Spending control | Per account, little visibility | Provider dashboard | Group quotas, per-key limits, reservation before every request |
| Where the data goes | On the provider's servers | On the provider's servers | You decide per model: cloud or local runtime, optional PII |
| Choice of models | The provider's | Provider catalogue | DeepSeek, OpenRouter and your own runtimes, with stable aliases |
| Model stability over time | Changes without notice | Depends on the provider | Stable aliases; on local runtimes you govern the weights |
| Continuity during an outage | You stop with them | You stop with them | With several equivalent targets and failover routing |
| Audit and accounting | Not provided | Provider reports | Balanced ledger, activity log, export |
| Multiple companies and resale | No | No | Tenants, resellers and commissions |
| Infrastructure | Not needed | Not needed | Your own Kubernetes cluster; GPUs only for local models |
How to size itNot a meter.
Not a meter.
A cluster to size.
Sizing starts from the people and applications working together and the models you want to serve, not from tokens. Each VM of a managed deployment hosts one complete replica of the model.
- Installation
- Four dedicated namespaces. Customer companies are application tenants: they get neither a cluster nor a database of their own.
- Capacity
- Add replicas or equivalent targets; routing distributes with leastBusy, weighted or failover.
- SLA
- No SLA unless agreed in writing: availability depends on the cluster, models and providers you choose.
- Idle time
- With authorised autoscaling, managed deployments stop and restart on demand. While starting, the gateway answers 503 model_starting with a Retry-After of 15 seconds.
01
02
03
04
How Ventic makes it possibleOpen models are cheap.
Open models are cheap.
Ventic makes them usable at work.
- One endpoint, unchanged clients
Coding agents and apps keep their OpenAI or Anthropic SDKs. Changing model never touches the code.
- Frontier only when needed
Rules by team, request type, PII and budget: that is how 90% of traffic goes to open models.
- GPUs and servers from one manifest
Rent GPUs or bring up the local server with venticli apply. Autoscaling and off-hours shutdown included.
- Local servers with no open ports
The agent connects outbound: no public IP, no VPN. The office server is a target like any other.
- Sensitive data stays in
Reversibly masked PII and residency rules: requests with personal data stay local.
- Costs in check, never down
Budget reserved before every request. If a target is slow or down, the rule moves to the next one.
Local inference: from PC to rack
Ryzen AI Max+ 495 with 192 GB of unified memory: open models for a developer team, under the desk.
Multi-GPU nodes in your datacenter for the whole company: Ventic manages them as one compute pool, with autoscaling.
Ventic Tuning · optional
The best setups for your hardware, always up to date.
We give you inference configurations already optimised for your machines, so you get the most out of every GPU. For the latest open-weight models we fine-tune the serving parameters on the main inference engines: quantization, parallelism, batching, KV cache and context.
- New open-weight models ready as soon as they ship
- vLLM, SGLang, llama.cpp and Ollama
- More tokens per second, more users per machine
€59.99per month per hardware setup
One fee for each distinct hardware configuration you want covered. The service is optional: Ventic works without it too.
Local AI server · sold by Compud
Ryzen AI Max+ PRO 495, 192 GB. Ready for Ventic.
- Memory
- 192 GB unified, 273 GB/s
- Models
- Qwen3.8 Flash Next, GLM-5.3 Flash, DeepSeek V4 Flash, gpt-oss-120b
- Speed per user
- 12–55 tok/s depending on model
€7.499
Manufacturer list price, VAT to be confirmed. About 6 developers per server; payback in 17 months vs the $100 plan.
Request the server from CompudWorth keeping in mind
- 01Monthly licence with a one-month commitment. Price list changes apply to new requests; subscriptions already agreed keep their amount.
- 02Hardware, hosting, AI consumption, SMTP, S3 storage, payments and third-party licences are separate. The demo includes no AI credit.
- 03Initial setup on request — you can also install it yourself with the guide. Later support by the hour, when needed.
- 04Prices exclude VAT. SLA only if agreed in writing.
Let's start with the analysis.
One hour to understand what you want to govern, which models you need and where they must run.