One-click install
Model recommendation from host RAM and disk.
Local AIBeta
An AI assistant, on your own server
Install an open language model on one of your servers and talk to it about your infrastructure; answers are generated entirely locally.
- 100%
- local processing
- Ollama
- open-source runtime
- 4 models
- Qwen and Gemma
Live
HOW IT WORKS
How it works
- 1
Set a RAM budget
Decide how much memory the model may use.
- 2
Get a recommendation
The console suggests the best server and model for the free resources.
- 3
Install with approval
The agent installs Ollama and the model; progress shows in the console.
- 4
Chat and analyze
Ask questions in the chat page or run a full organization analysis.
Capabilities
Local AI
Chat in the console
Ask about servers, CVEs and upgrades.
Periodic analysis
Automatic read-only reports.
No data egress
Ollama binds to loopback only.
No auto-execution
Nothing runs without admin approval.
Chat history
Conversations stay in your organization.
USE CASES
Use cases
Ask about your infrastructure
"Which servers have critical CVEs?" answered from real data.
Security prioritization
Summaries of vulnerabilities and a suggested fix order.
Log and incident analysis
Explain errors and suggest next steps without sending logs out.
Documentation
Generate explanations and runbooks from the current configuration.
Technical specs
- Runtime
- Ollama in a container, loopback only
- Models
- Qwen 3.5 4B · Qwen 3 8B · Qwen 3.5 9B · Gemma 4 12B
- Minimum RAM
- 6 GB free + 2 GB reserved for the OS
- GPU
- Optional; used when present
- Transport
- Through whyconnect; no public port
- Actions
- Only after admin approval, one at a time
Comparison
| whycloud local AI | Public AI service | |
|---|---|---|
| Where data is processed | Your server | External servers |
| Logs and config leave the network | No | Yes |
| Per-token cost | None | Yes |
| Works offline | Yes | No |
PRIVATE AI
Privacy by design
Telemetry is passed to the model as untrusted data and model output is never turned into commands.
- Qwen and Gemma models
- Container memory limits
- Runs only with install permission
FAQ
Frequently asked questions
Is our data used for training?
No; the model runs on your server.
What hardware is required?
At least 6 GB of free RAM; GPU is optional.
Can the AI execute commands?
No; it only suggests, and allowed actions need admin approval.
Can the model reach the internet?
No. Ollama listens on the host loopback only and requests travel through the agent.
Can the model run commands by itself?
No. Model output is a suggestion; every action needs explicit admin approval.
Does it work without a GPU?
Yes. Smaller models run on CPU, just more slowly.
Related services
Live streamingwhycloud live streaming: RTMP/SRT ingest, adaptive transcoding, record…Kafka as a ServiceManaged Kafka by whycloud: clusters, topics, partitions, consumer lag …Video & audio on demandwhycloud media library: upload or import, automatic renditions, subtit…Object StorageConnect whycloud to any in-house S3 (MinIO, Ceph…) without a public IP…
Turn on your own cloud today
Create an account and connect your first server in under five minutes.