Noema
    Noema

    Remote Endpoints

    Release 1.3 introduces remote endpoints, letting Noema connect to inference servers that live on desktops, lab machines, or cloud VMs while you keep the familiar chat interface on your phone or tablet.

    Eligibility

    Make sure every device runs Noema 1.3+

    Remote registration requires version 1.3 or later on both the controlling device and the machine hosting the backend. Earlier builds lack the authentication flow and catalog support that remote backends rely on.

    Overview

    What a remote backend stores

    • Friendly name that appears throughout the library.
    • Base URL plus chat and model catalog paths tailored to each provider.
    • Optional authorization header for secure deployments.
    • Provider type (OpenAI API, LM Studio, Ollama, or custom HTTP).
    • 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 spot in the chat picker.
    3. Enter the base URL; Noema warns if you target a loopback host or if Ollama is not exposed via OLLAMA_HOST=0.0.0.0.
    4. Choose a provider preset to auto-fill chat and model routes.
    Endpoints & auth

    Configure routes and headers

    Provider presets stay in sync as you switch between OpenAI API, LM Studio, and Ollama, but you can override any field with an absolute URL or relative path. Expand the authentication drawer to capture headers such as Authorization: Bearer sk-…; Noema will send them with every request.

    Monitoring

    Track connection health

    After you save, Noema fetches the remote catalog and records the HTTP outcome in a connection summary. Status chips in both 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.

    Custom models

    Augment the catalog manually

    Add bespoke identifiers whenever a server omits certain deployments. They deduplicate against fetched entries and fold back into the picker so your one-off builds remain selectable.

    Use a remote model

    Activate and chat

    • Status pills show availability, quantization, and format at a glance.
    • The Use action warms the backend (unless Off-Grid mode blocks calls) and switches you directly into chat.
    • Provider extras—like OpenAI tool schemas or Ollama metadata—display in the detail view so you can tailor prompts.