Offline Operation & Privacy FAQ
Noema is built for privacy-first, offline-first use. These answers explain what runs locally, when network activity occurs, and how to troubleshoot common questions.
Connectivity
Does Noema need the internet?
No. After installing and downloading at least one model, all core chat and dataset features run entirely on-device. Go offline anytime; only optional features like remote endpoints, web search, or new model downloads require a connection.
Network usage
What data ever leaves my device?
- Model downloads pull binaries from trusted hubs like Hugging Face.
- Dataset imports fetch from the original source when you choose to download them.
- Web search sends only the generated query to our private SearxNG relay and processes results locally.
- Version checks ping Noema’s update endpoint—no prompts or transcripts are uploaded.
Protections
How is my data safeguarded?
- Inference runs locally, so prompts and outputs never reach external servers.
- Noema works without an account and collects no analytics or telemetry.
- The app is built on auditable open-source components like llama.cpp.
- All assets live inside the iOS sandbox—protected by your device passcode and Secure Enclave.
Limits
What are the trade-offs of running locally?
Mobile hardware is slower than data center GPUs. Expect a few tokens per second on 4B–7B models and plan for smaller quantizations on devices with limited RAM. The upside: no quotas, no rate limits, and full control over your data.
Troubleshooting
Common questions
- Model will not load: free up RAM by closing apps or choose a lighter quantization.
- Inaccurate answers: upgrade to a larger model, refine your prompt, or enable web search/datasets for extra context.
- Need unlimited web lookups? You're set—SearxNG runs without caps, so confirm the search toggle is on and start exploring.