# NoemaAI

> NoemaAI researches small language models, releases them as open weights, and builds Noema, a private AI app for iPhone, iPad, Mac and Apple Vision Pro.

NoemaAI publishes its models as open weights on Hugging Face (https://huggingface.co/NoemaAI-labs) and its documentation as Markdown. Add `.md` to any documentation URL for the page source, use https://noemaai.com/docs/index.md for the full index, or read everything at https://noemaai.com/llms-full.txt.

## Models

- [Noema 1.5 4B Deep](https://noemaai.com/models/noema-1-5-4b#deep): Thinks longer on hard problems. The more accurate of the two 4B styles. Weights: https://huggingface.co/NoemaAI-labs/Noema-1.5-4B-Deep. Open weights. Released October 13, 2026.
- [Noema 1.5 4B Fast](https://noemaai.com/models/noema-1-5-4b#fast): Answers in a fraction of the tokens and almost never gets stuck. Suits phones. Weights: https://huggingface.co/NoemaAI-labs/Noema-1.5-4B-Fast. Open weights. Released October 13, 2026.
- [Noema 1.5 2B](https://noemaai.com/models/noema-1-5-2b): A compact model with better instruction following and shorter reasoning. Weights: https://huggingface.co/NoemaAI-labs/Noema-1.5-2B. Open weights. Released August 12, 2026.
- [Noema 2B](https://noemaai.com/models/noema-2b): NoemaAI's first model. Deprecated: use Noema 1.5 2B instead. Weights: https://huggingface.co/NoemaAI-labs/Noema-2B. Open weights. Released July 6, 2026.
- [Model reference](https://noemaai.com/docs/models.md): every repository, format, license and release date in one table.
- [Run Noema models](https://noemaai.com/docs/run-noema-models.md): commands for llama.cpp, Ollama, LM Studio, MLX and Transformers.

## Research

- [Reusing Model Probabilities for New Queries](https://noemaai.com/research/reusing-model-probabilities): Can one set of model probabilities answer many different questions? This paper studies how fine-grained predictions can be combined into broader categories, keeping answers consistent across label groupings without querying a model again for each grouping. Preprint, October 23, 2026, https://doi.org/10.5281/zenodo.23584994.

## Docs: App

- [How Noema works](https://noemaai.com/docs/about-noema.md): What local-first means, what can stay offline, and when Noema uses a network connection.
- [Installation](https://noemaai.com/docs/installation.md): Current system requirements, App Store installation, and first-model setup.
- [Quick start](https://noemaai.com/docs/quick-start.md): Install Noema, choose a starter model, and send your first fully local message.
- [Blueprints](https://noemaai.com/docs/blueprints.md): Review a proposed model, runtime, and privacy setup on the actual device before applying it.
- [Chat interface](https://noemaai.com/docs/chat-interface.md): Keep a conversation moving, compare answers, and find or reuse earlier work.
- [Deep Research](https://noemaai.com/docs/deep-research.md): Review a research plan, then let Noema search in steps and write a report with numbered citations.
- [Datasets and RAG](https://noemaai.com/docs/dataset-integration.md): Supported imports, embeddings, Hugging Face datasets, retrieval settings, and Dataset Health.
- [Dataset indexing](https://noemaai.com/docs/dataset-indexing.md): How embedding works, why the time estimate can start as …, and what to check when indexing stalls.
- [Knowledge Packs](https://noemaai.com/docs/knowledge-packs.md): Curated, license-cleared offline collections for local retrieval.
- [Voice and media](https://noemaai.com/docs/voice-and-media.md): Speech recognition, spoken responses, Voice Mode, and media transcription.
- [Running models locally](https://noemaai.com/docs/running-llms.md): How GGUF, MLX, ExecuTorch, Core ML, and Apple Foundation Models differ, including the Noema 1.5 4B recommendations in 4.2.
- [Download and install models](https://noemaai.com/docs/downloading-models.md): From Explore to Stored: formats, fit guidance, the download queue, installation status and recovery.
- [Model settings](https://noemaai.com/docs/model-settings.md): A practical guide to presets, context, sampling, helper models, and format-dependent controls.
- [Reasoning history](https://noemaai.com/docs/reasoning-history.md): Decide, per model, whether earlier reasoning stays in context for follow-up turns.
- [Noema Overfit](https://noemaai.com/docs/noema-overfit.md): Run compatible Qwen, Gemma, Laguna, and DeepSeek mixture-of-experts models using a resident core and locally paged expert weights, then measure whether the model is practical on your device.
- [Chat tools](https://noemaai.com/docs/tool-usage.md): One remembered switch per tool, shared by the Tools tab, Settings, and every chat on this device.
- [Web Search](https://noemaai.com/docs/web-search.md): Search controls, source reading, custom SearXNG behavior, and privacy boundaries.
- [Python code execution](https://noemaai.com/docs/python-code-execution.md): Run Python locally inside Noema for calculations, data analysis, machine learning, symbolic mathematics, graph analysis, and scientific visualization.
- [Meeting transcription](https://noemaai.com/docs/meetings.md): Turn a recording into a transcript you can review, summarize, and ask questions about.
- [Translation](https://noemaai.com/docs/translation.md): Download a translation model once, then translate among 38 languages without sending the text to a remote AI service.
- [Autopilot](https://noemaai.com/docs/autopilot.md): Per-message routing, Phone a Friend, escalation policy, privacy, fallbacks, and receipts.
- [Constellation](https://noemaai.com/docs/constellation.md): Your chats follow you, and your Mac’s models become available across your personal devices.
- [Projects and Remote Control](https://noemaai.com/docs/projects-and-remote-control.md): Carry project context everywhere while folders and coding tools stay on the Mac that owns them.
- [Constellation models, tools and photos](https://noemaai.com/docs/constellation-models-and-tools.md): The model runs on the Mac while the requesting device keeps control of tools, local knowledge, permissions, and the chat.
- [Datasets shared by your Mac](https://noemaai.com/docs/mac-shared-datasets.md): Retrieve passages from datasets your Mac shares; the original files and indexes stay on the Mac.
- [Constellation Boost](https://noemaai.com/docs/constellation-boost.md): Let your iPhone help a local model on your Mac process long prompts or hold more context.
- [Constellation connections and privacy](https://noemaai.com/docs/constellation-connections.md): Noema tests the available paths when you press Use and selects the best working route before the first message.
- [Constellation troubleshooting](https://noemaai.com/docs/constellation-troubleshooting.md): Check both devices, confirm the Mac is awake, inspect the selected route, and use the ejection controls to manage Mac memory.
- [Mac projects and Agent](https://noemaai.com/docs/mac-projects-and-agent.md): Keep project context together and choose exactly how Noema may inspect or change a workspace.
- [Mac tools and benchmarks](https://noemaai.com/docs/mac-tools-and-performance.md): The toolbar memory readout, Tools navigation, and Benchmarking Center results explained.
- [Privacy and network activity](https://noemaai.com/docs/privacy-and-network-activity.md): A feature-by-feature map of local, LAN, iCloud, Apple, Noema-operated, and third-party data paths.
- [Support](https://noemaai.com/docs/support.md): What to include in a report and where to find logs, diagnostics, and contact controls.

## Docs: Models

- [Noema models](https://noemaai.com/docs/models.md): Every released Noema model with its repository, size, base model, formats, license and release date.
- [Model licenses](https://noemaai.com/docs/model-licenses.md): Every Noema model is an open-weight release, not open source, and where each set of terms lives.
- [Run Noema models](https://noemaai.com/docs/run-noema-models.md): Copyable commands for llama.cpp, Ollama, LM Studio, MLX and Transformers, with the file each model card recommends.
- [Prompt format and sampling](https://noemaai.com/docs/prompt-format.md): Chat template, thinking and non-thinking modes, and the sampling and output limits each model card recommends.
- [Evaluation methodology](https://noemaai.com/docs/evaluation-methodology.md): Paired comparisons with each model's base, identical prompts and graders, stated presets and limits, and 95% intervals.

## Docs: Developers

- [Local API quickstart](https://noemaai.com/docs/local-api-quickstart.md): Start the Local API on your Mac, copy its address and key, and send your first request.
- [Local API reference](https://noemaai.com/docs/local-api.md): Let compatible apps and coding clients call models hosted by Noema on your Mac.
- [MCP connectors](https://noemaai.com/docs/connectors.md): Give a compatible model tools from a server you choose, with approval before it runs them.
- [Remote endpoints](https://noemaai.com/docs/remote-endpoints.md): Provider disclosure, endpoint security, supported routes, credentials, and Off-grid behavior.

## Docs: Teams

- [Noema Teams administration](https://noemaai.com/docs/teams-admin.md): What each area of the Teams console controls, and what employees need to join.

## Products

- [Noema app](https://noemaai.com/app): Private AI for iPhone, iPad, Mac and Vision Pro that runs models on hardware you control.
- [Download Noema](https://noemaai.com/download): The App Store and Direct Mac downloads for each platform.
- [Noema for Mac](https://noemaai.com/mac): The Direct Mac edition and how it differs from the App Store edition.
- [Constellation](https://noemaai.com/constellation): Use your Mac's models from iPhone and iPad.
- [Overfit](https://noemaai.com/overfit): Run mixture-of-experts models larger than memory.
- [Noema Search](https://noemaai.com/noema-search): Private web search with sources you can check.
- [Release notes](https://noemaai.com/changelog): What changed in each version of the app.
- [Noema Teams](https://noemaai.com/enterprise/teams): Enroll employees, set role policy and manage approved devices from one console.
- [Company knowledge](https://noemaai.com/enterprise/teams#knowledge): Answer from approved internal documents. Original documents, private indexes and decryption keys stay in your environment and on approved devices.
- [Custom model pilots](https://noemaai.com/enterprise/custom-models): Adapt a model to your terminology and formats, measured on your task and hardware. Offered as a pilot.

## Developers

- [Developers](https://noemaai.com/developers): Run Noema models, the Local API on Mac, and open-source tools.
- [Noema Atlas](https://noemaai.com/developers/atlas): Verified, peer-to-peer distribution for open model weights. Open source.
- [Hugging Face](https://huggingface.co/NoemaAI-labs): Noema model weights and model cards.
- [GitHub](https://github.com/noemaai-labs): Source code for Noema Atlas and other NoemaAI projects.

## Optional

- [News](https://noemaai.com/news): Model releases, product updates and company news.
- [News feed](https://noemaai.com/news/rss.xml): RSS.
- [Research feed](https://noemaai.com/research/rss.xml): RSS.
- [Company](https://noemaai.com/company): What NoemaAI works on and who we are.
- [Trust](https://noemaai.com/trust): Architecture, data handling and disclosure.
- [Security](https://noemaai.com/security): Security contact and leadership.
- [Privacy policy](https://noemaai.com/privacy)
- [Terms of service](https://noemaai.com/terms)
- [Sitemap](https://noemaai.com/sitemap.xml)