AI Analyst / Local AI

Local AI — Maximum Privacy

AVAILABLE NOW We'll choose a sensible local model for this computer instead of making the user guess.

← AI Analyst
🔒 Maximum privacy. Your CRM workbook stays in the browser. Only the compact analytics context selected for an AI question is sent to Ollama running on this computer.
Local setup wizard

Set up your private AI Analyst

The wizard detects what it can, asks for RAM only when the browser cannot reliably read it, and recommends an official Qwen 3.5 model conservatively.

Step 1 of 4
1
Check this computer
We can identify your operating system and browser CPU concurrency. Desktop browsers generally do not expose exact RAM reliably, so we may ask you.
OSChecking…CPU threadsChecking…RAMNot available
Checking your browser capabilities…
2
Get your recommended model
You do not need to compare quantization files or model variants. CRM Intelligence will select one official Qwen 3.5 tag.
Complete Step 1 first
Your recommendation will appear here.
Conservative model guide
qwen3.5:0.8b
~1.0 GB · very low resource usefallback
qwen3.5:2b
~2.7 GB · lightweight8 GB
qwen3.5:4b
~3.4 GB · balanced16 GB
qwen3.5:9b
~6.6 GB · stronger24–32 GB
qwen3.5:27b
~17 GB · high resource use64 GB+
3
Install Ollama & allow CRM connection
Ollama is the local runtime. Install it first, then complete the one-time browser connection step below so this website can talk to Ollama on this computer.
Complete the computer check to see the operating-system-specific instructions.
Complete the Ollama installation and one-time connection permission above before installing the recommended model.
4
Connect & open AI Analyst
Once Ollama is running and the recommended model exists locally, we verify the connection and open the dedicated chat page.
ConnectionNot checked
Detected model—
Run the connection check after Ollama is open.
Privacy boundary: CRM Intelligence never needs your Ollama password or API key. Your Excel/CSV stays in the browser. The AI request goes to localhost:11434 on the same computer.
Model recommendations are conservative application guidance, not official hardware requirements. A larger model can increase RAM/CPU/GPU use, so this wizard intentionally prefers the smallest model likely to be useful for the available memory.

Why do we ask about RAM?

Local AI quality and resource use depend heavily on model size. Instead of asking a customer to choose among many Qwen variants, this wizard turns a simple RAM choice into a practical recommendation.

If the customer selects Not sure, the wizard uses a lightweight fallback and does not assume a high-end computer.