Noema app · by NoemaAI
A private AI assistant for iPhone, iPad, Mac and Vision Pro.
Noema runs language models on your own devices, answers from your files, searches the web with sources you can check and works in projects on your Mac. Personal local use needs no account, and Off-Grid Mode stops Noema's online features whenever you choose.
See it work
One app for models, files, web search and projects.
Captures from the Noema app. Choose a tab to see each feature.
Download open models, including NoemaAI's own, and see which ones fit your device before you load them.Downloading models
Capabilities
What you can do with Noema.
Chat with local models
Run GGUF, MLX, Core ML and ExecuTorch models, or Apple's on-device Foundation Model, in one native app. Local chat keeps working offline.
Running models locallyAsk your files
Add PDFs, EPUBs, Markdown, text and recordings, index them on your device and get answers with the source passages.
Datasets and retrievalDeep Research and web search
Noema Search is built in with no setup. Search the web from a chat, or review a research plan and get a report with numbered citations.
Noema SearchProjects and Agent on Mac
Keep a folder, instructions and sources in one project. Agent proposes edits and commands; you approve them and review the changes.
Noema for MacAcross devices with Constellation
Chats and projects sync through your private iCloud, and iPhone, iPad or Vision Pro can use the models on your awake Mac.
ConstellationLarger models with Overfit
Run compatible mixture-of-experts models whose full weights don't fit in memory, with experts read from local storage.
Overfit
Runs on Noema models
Open-weight models from the same lab.
NoemaAI post-trains small open models and releases the weights. In Noema 4.2, Noema 1.5 4B appears first in Explore: Deep for longer math and coding reasoning, Fast for brief everyday reasoning. You can also run any compatible open model.
Noema 1.5 4B Deep
Thinks longer on hard problems. The more accurate of the two 4B styles.
GPQA Diamond +15.6 points over Qwen3.5-4B, under Qwen's own sampling preset
Model cardNoema 1.5 4B Fast
Answers in a fraction of the tokens and almost never gets stuck. Suits phones.
Repetition loops on hard prompts fall from 65.0% to 1.1% (greedy decoding)
Model cardNoema 1.5 2B
A compact model with better instruction following and shorter reasoning.
IFEval prompt-strict +7.0 points over stock Qwen3.5-2B (72.1 vs 65.1)
Model card
Each figure is a gain over the model's own base under the stated setting. Methods, confidence intervals and limits are on each model page. All Noema models
Platforms
Where each feature runs.
Taken from the platform list and version on each documentation page. A dash means the guide does not list that device.
| Feature | iPhone | iPad | Mac | Vision Pro | Guide version |
|---|---|---|---|---|---|
| Chat and models | |||||
| Local modelsNoema 4.2+ | Yes | Yes | Yes | Yes | 4.2+ |
| Side Question, Compare Models, BranchNoema 4.2+ | Yes | Yes | Yes | Not listed | 4.2+ |
| Autopilot routingNoema 3.5+ | Yes | Yes | Yes | Yes | 3.5+ |
| Overfit (running packages)Noema 3.6+ | Yes | Yes | Yes | Yes | 3.6+ |
| Files and knowledge | |||||
| Datasets and retrievalNoema 4.2+ | Yes | Yes | Yes | Yes | 4.2+ |
| Knowledge PacksNoema 4.2+ | Yes | Yes | Yes | Yes | 4.2+ |
| Web | |||||
| Web searchNoema 4.2+ | Yes | Yes | Yes | Yes | 4.2+ |
| Deep ResearchNoema 4.2+ | Yes | Yes | Yes | Not listed | 4.2+ |
| Tools and voice | |||||
| PythonNoema 4.2+ | Yes | Yes | Yes | Yes | 4.2+ |
| MCP connectorsNoema 4.2+ | Yes | Yes | Yes | Not listed | 4.2+ |
| Voice Mode and mediaNoema 3.1+ | Yes | Yes | Yes | Yes | 3.1+ |
| Meeting transcriptionNoema 4.2+ | Yes | Yes | Yes | Not listed | 4.2+ |
| Across devices | |||||
| ConstellationNoema 4.2+ | Yes | Yes | Yes | Yes | 4.2+ |
| Constellation BoostNoema 4.1+ | Yes | Not listed | Yes | Not listed | 4.1+ |
| Projects and Remote ControlNoema 4.2+ | Yes | Yes | Yes | Not listed | 4.2+ |
| Mac | |||||
| Projects and Agent workbenchNoema 4.2+ | Not listed | Not listed | Yes | Not listed | 4.2+ |
| Local API serverNoema 4.2+ | Not listed | Not listed | Yes | Not listed | 4.2+ |
Overfit packages run on every device; creating a package from a GGUF needs a Mac. The Direct Mac edition adds terminal commands, Git review and local MCP servers. Compare Mac editions
Privacy
Local by default. Online only when you choose.
Local models, documents, indexes and conversations can stay on your device. Optional features say where they connect, and Off-Grid Mode stops them.
| Feature | Destination | What may be sent |
|---|---|---|
| Local models and RAG | Your device | Nothing leaves for generation or local retrieval. |
| Local Python, Memory, Calculator, Converter, Charts | Your device | Tool inputs and results remain local; Python has no network access. |
| Web Search / Deep Research | Selected search service and source reader | Queries, locale, controls, candidate URLs or snippets, source references, and open/find parameters; a remote report-writing model also receives the relevant findings. |
| Constellation chat sync | Private iCloud database | Chat and project payloads; models, datasets, and source files remain device-local. |
| MCP connectors | Server you configure | Approved tool-call arguments and returned results. Remote connectors use the network even with a local chat model. |
| Remote endpoint | Configured provider | Prompt, relevant chat and system context, supported attachments, tools and results, and retrieved excerpts. |
Off-Grid Mode blocks external HTTP and HTTPS traffic from Noema. Local generation, retrieval and tools keep working. Read the full network activity guide
Free on the App Store.
One download for iPhone, iPad, Mac and Apple Vision Pro. Configure a project on the web first with Blueprint, if you like.