AI inside your CRM saves real time: summarise tickets, score leads, generate a Customer 360 analysis at the click of a button. But two things often hold small teams back. First, OpenAI or Claude charge per token — and that adds up. Second, there is the question of where your customer data actually goes. Here is the reassurance up front: the AI Suite anonymises personal data before it leaves your system — so your data protection holds regardless of the provider. More on that below.

The good news: AI in your CRM can also be free. With OpenRouter and a free NVIDIA Nemotron model, the AI Suite runs at no per-token cost. And if you want full control over your data, you can even run Nemotron on your own server. This post shows the setup — plus the two pitfalls we found and solved along the way.

Quick start — for a fast start and good results
Provider: Custom (OpenRouter) · Model: nvidia/nemotron-3-nano-30b-a3b:free · Important: enable „Disable reasoning (for JSON tasks)“.
For deeper analyses (Customer 360, Lead Summary): nvidia/nemotron-3-super-120b-a12b:free — also with „Disable reasoning“.
These are the three settings that turn „free“ into „actually works“. Why exactly these — the rest of this post explains.

Why use a free model at all?

  • No running costs. Free models do not charge per token. You use every feature without watching a usage budget.
  • Test risk-free. Before committing to a paid provider, you can take your time to see whether the AI meets your needs.

Why Nemotron — and not just any free model

OpenRouter’s catalogue of free models is large and changes constantly. We recommend NVIDIA Nemotron because it combines several things:

  • Stably hosted — it does not disappear overnight like many preview models.
  • Multilingual — the output follows your profile language, including German, Italian, French, Spanish and Arabic.
  • Clean, structured results — especially for JSON-based features like lead scoring and sentiment.
  • Open-licensed — and that is the decisive point for data protection (more below). You can self-host Nemotron. With OpenAI or Claude that is simply not possible.

What about data protection?

Important up front: your data protection stays intact — through the AI Suite’s anonymisation. Before a prompt leaves your system, the Privacy Shield removes or pseudonymises personal data such as names, email addresses and phone numbers. So no clear-text personal data goes to the AI provider — whether that is OpenAI, Claude or Nemotron.

Still, an honest distinction, because „NVIDIA“ does not automatically mean „EU“:

  • Via OpenRouter: inference runs on NVIDIA / third-party infrastructure, typically not in the EU. You are protected by the anonymisation above — the personal data stays with you.
  • Self-hosted: because Nemotron is open-licensed, you can run it via Ollama on your own server (for example in the EU). Then the data never leaves your system at all — full data sovereignty.

In short: free you get right away. Data sovereignty does not come „for free through Nemotron“ — it comes from the AI Suite’s anonymisation and, if you need it, from self-hosting.

What we learned

Two things were not obvious:

1. „openrouter/free“ is a trap

It seems natural to just enter the auto-router openrouter/free. In practice it is a placeholder that picks a different model per request — with unpredictable results. In our tests it sometimes returned a short canned phrase instead of an analysis. The lesson: pin a real, fixed model rather than a router.

2. Reasoning models „overthink“ their token budget

Modern models like Nemotron „think“ internally before answering. On tasks with a tight budget — lead scoring and sentiment run on very few tokens — that budget is used up before an answer arrives. The result: nothing usable, with no visible error.

Our fix has been built into the module since AI Suite 3.0.3: the „Disable reasoning (for JSON tasks)“ toggle. With it on, the model answers directly — fast and reliable, even within tight limits. This single switch is what turns „free in theory“ into „works in practice“.

How to set it up

  1. Create a free OpenRouter account and generate an API key.
  2. Go to Setup → AI Suite → Provider and choose Custom (OpenRouter). Enter the URL https://openrouter.ai/api/v1 and your key.
  3. Set the model to nvidia/nemotron-3-nano-30b-a3b:free.
  4. Enable „Disable reasoning (for JSON tasks)“. Done.

Tip: for deeper analyses like Customer 360 or Lead Summary, the larger nvidia/nemotron-3-super-120b-a12b:free is worth it — also with „Disable reasoning“ enabled.

An honest assessment

The first results are genuinely strong: clean, structured analyses in the respective language. Three points we keep transparent:

  • We are still running longer-term tests before recommending Nemotron as a fixed default.
  • For the Revenue Forecast feature, where numerical precision matters, a paid model can be the better choice.
  • Free tiers have rate limits — no problem for calm single use, but noticeable at high volume.

Bottom line: with Nemotron and the „Disable reasoning“ toggle, you get capable AI for Perfex CRM — at no per-token cost, and with a clear path to full data sovereignty when you need it.

Try it directly in our live demo, or take a look at the AI Suite.