Noema
    Noema
    Remote Models

    Remote Endpoints

    Remote endpoints let Noema connect to inference servers that live on desktops, lab machines, or cloud VMs while you keep the same library and chat flow on your Apple device.

    Want your phone to use the model running on your Mac without manual backend setup? Turn on Noema Relay and let devices discover each other automatically. Read the quick guide in Noema Relay.

    OpenAI API

    Good for OpenAI-compatible /v1 backends.

    OpenRouter

    Good when you want a large hosted catalog and provider routing.

    LM Studio

    Good for LM Studio’s local or hosted REST server.

    Ollama

    Good for simple local or LAN-hosted Ollama usage.

    Noema Relay

    Good for pairing another device with your Mac and using the provider configured there.

    Eligibility

    Use current builds on both sides

    Remote registration depends on current backend presets, authentication flow, and catalog handling. Make sure the Noema device you are using and the machine hosting the backend are both on recent builds before troubleshooting connectivity.

    Overview

    What a remote backend stores

    • Friendly name shown in the library and chat picker.
    • Base URL plus provider-specific chat and model catalog routes.
    • Optional authorization header for secured deployments.
    • Provider type: OpenAI API, OpenRouter, LM Studio, Ollama, or paired Noema Relay behavior.
    • Custom model identifiers that merge with the fetched catalog.
    Add a backend

    Register a server from the library

    1. Open Stored and press the + button.
    2. Name the server so it is easy to recognize in the chat picker.
    3. Enter the base URL, then choose the provider preset to auto-fill model and chat routes.
    4. Add headers when required, such as Authorization: Bearer ....
    5. Save and let Noema fetch the remote model catalog.
    OpenRouter endpoint

    Add OpenRouter as a hosted remote provider

    Use OpenRouter when you want access to OpenRouter's hosted model catalog and provider network instead of connecting to a single local server.

    How to add it
    1. Open the remote backends screen and choose Add Custom Backend.
    2. Set Endpoint Type to OpenRouter.
    3. Enter a name, or keep the default OpenRouter.
    4. Leave the base URL as https://openrouter.ai. Noema fills this in automatically.
    5. Paste your OpenRouter API key. You do not need to provide a custom auth header.
    6. Optionally add one or more model identifiers if you want to pin specific models manually. Otherwise, save first and load the catalog afterward.
    7. Save the endpoint. Noema verifies the API key before completing setup.
    What Noema handles automatically
    • Uses OpenRouter's standard Bearer API-key authentication.
    • Uses the default OpenRouter chat and model endpoints.
    • Fetches the OpenRouter model list for the account.
    • Adds OpenRouter attribution headers automatically.
    After setup
    • Browse and search the OpenRouter model catalog.
    • Filter models by capabilities such as tools, reasoning, vision, structured outputs, moderation, pricing, and supported parameters.
    • Favorite models for quicker access.
    • Open model info to inspect metadata like context length, provider details, pricing, and supported parameters.
    • Open per-model settings and adjust context length, temperature, top-p, top-k, min-p, and repetition penalty.
    OpenRouter-specific behavior
    • If OpenRouter publishes default sampling values for a model, Noema can reset that model back to those defaults.
    • Noema only sends sampling and tool parameters that the selected OpenRouter model reports as supported, which helps avoid invalid requests.
    Good fit

    OpenRouter is the best remote option when you want a large hosted model catalog, provider routing, and per-model capability-aware settings without manually wiring a separate inference server.

    Provider reference

    When to use each remote type

    OpenAI API

    Use this when you are targeting a standard OpenAI-compatible /v1 backend. It is the broadest preset for cloud or self-hosted services that follow the OpenAI request shape.

    OpenRouter

    Large hosted model catalog with provider routing.

    Per-model settings available
    Best for

    Use OpenRouter when you want a broad hosted catalog and provider routing without manually wiring separate endpoints for every hosted model.

    What Noema exposes

    Per-model remote settings are available, including context length and standard sampling controls like temperature, top-p, top-k, min-p, and repetition penalty. Models can also surface provider defaults.

    LM Studio

    Connect to LM Studio’s local or hosted REST server.

    Per-model settings available
    Best for

    Use LM Studio when you want Noema to connect to a machine that already serves models over LM Studio’s REST interface.

    What Noema exposes

    LM Studio models expose a settings sheet with context length plus standard sampling controls such as temperature, top-p, top-k, min-p, and repetition penalty.

    Ollama

    Use Ollama when you want simple local or LAN-hosted model serving and easy model pulls. Expose the server properly if you need Noema to reach it from another device.

    Noema Relay

    Use Noema Relay when you want to pair another device with your Mac and use the provider configured there. It is the least manual route when you are already inside the Noema ecosystem.

    Endpoints & auth

    Configure routes and headers

    Provider presets help fill chat and catalog routes for OpenAI API, OpenRouter, LM Studio, and Ollama, but you can still override fields with absolute URLs or relative paths. Use the authentication drawer for headers such as Authorization when the provider requires them.

    Monitoring

    Track connection health

    After you save, Noema fetches the remote catalog and records the HTTP outcome in a connection summary. Status chips in the list and detail views mirror that state, and a manual reload control lets you refresh on demand. When Off-Grid mode is active, network calls pause until you reconnect.

    Remote model settings

    Which providers currently expose per-model controls

    Remote per-model settings are currently centered on LM Studio and OpenRouter. In both cases, Noema can expose context length and standard sampling controls like temperature, top-p, top-k, min-p, and repetition penalty.

    Custom models

    Augment the catalog manually

    Add custom identifiers whenever a server omits a deployment you still want to address. Noema deduplicates them against fetched entries and folds them back into the picker so one-off builds remain selectable.