Provider quickstart

This guide takes a self-hosted source from nothing to Live, when customers can route requests to it. It uses Ollama because it needs no API key; the steps are the same for every other source. Source-specific requirements are in Inference sources.

You need a Numes account, a machine on a supported platform, and an inference source that the machine can reach.

1. 1

   Create a provider (optional)

   Setup can create the provider for you, but creating it in the dashboard first lets you pick the name customers see. Open Provider › Overview, choose Become a provider (or New provider), enter the Provider name and choose Self-hosted under How you'll serve. The dialog shows the commands below and waits for your first endpoint to connect; you can close it and come back.

2. 2

   Install the agent

   Shell

   Copy Shell code

   curl -fsSL https://numes.ai/install | sh

   The installer downloads the latest stable release for your platform, verifies it, and links the numes command at ~/.local/bin/numes. If that directory isn't on your PATH, it prints the line to add. Check the installation:

   Shell

   Copy Shell code

   numes version

   To pin a version, install offline or control updates, see Install and update.

3. 3

   Sign in

   Shell

   Copy Shell code

   numes auth login

   At a desktop session, a browser opens on this machine. Over SSH, without a display or in a non-interactive terminal, the agent prints an address and a short code instead: open the address on any device, check that the code matches and approve. Force either method with --login-method loopback or --login-method device.

   numes auth status shows your sign-in, and numes auth logout revokes and removes it.

4. 4

   Create an endpoint

   Shell

   Copy Shell code

   numes provider tui

   Press n for a new endpoint. Setup has four steps:

   1. Source: if asked, choose the account and provider you created in the dashboard; the dashboard dialog names both. Pick Ollama, check the address and concurrency, and choose Connect and find models →.

   2. Models: tick models with Space and search with /. Rows marked ⚠ needs name need a public author/model name; press e to choose one.

   3. Review: check each model's public name, currencies, prices, context length and capacity, and the provider and endpoint names. Prices are prefilled where a reference price exists; enter or confirm every one.

   4. Submit: press P. Setup creates or reuses the provider and endpoint, enrolls this machine, submits the models and, with Start serving now on, starts the endpoint.

   Each step saves a draft. Esc or a tab key leaves setup, and Continue setup on Overview or Endpoints resumes it. See Terminal UI for every screen and key.

5. 5

   Wait for review and verification

   Your provider, your endpoint, each public model name and each model's first prices are reviewed. Meanwhile your models show In review or Waiting for approval.

   After approval, models show Checking while automated checks exercise every capability you declare through your agent, with synthetic requests and never customer data. When every check passes, the model shows Live and starts receiving traffic. Checks run only while your agent is connected, so keep the endpoint running.

   Later price or model changes go live automatically once they pass verification and price protection. Misrepresenting a model, its capabilities, its usage or its data handling leads to suspension and to withholding or reversal of earnings.

6. 6

   Check status

   Shell

   Copy Shell code

   numes provider status --profile ollama-local

   status combines the local process with the platform's view of your provider, endpoint and models, and reports one readiness state and the next action. Add --local to skip sign-in and check only the local process.

Set up from the command line

The same setup as commands you can script or run over SSH. Preview it with --plan, which saves a local draft and creates nothing on the platform:

numes provider setup ollama \  --profile ollama-local \  --provider-name "Example Labs" \  --model llama3.2:3b \  --canonical-model llama3.2:3b=meta-llama/llama-3.2-3b-instruct \  --currency USD \  --prompt-price 0.10 \  --completion-price 0.40 \  --plan --non-interactive

Then resume the draft, confirm the plan and start the endpoint:

numes provider setup --resume ollama-local --start --wait-ready

• --model is the model ID as your source reports it. Repeat it for each model.

• --canonical-model sets the public author/model name. Setup proposes a name for each model from names the platform already knows and from public model catalogues; you need this option when there's no proposal or you want a different name. Public model names are reviewed by the platform.

• Prices are exact decimals per 1M tokens in the chosen currency.

• --provider ID reuses a provider you created in the dashboard instead of --provider-name. Without either flag, interactive setup lists your self-hosted providers to reuse, then Create a new provider. Add --account ID if you belong to several accounts; numes account list shows them.

• --wait-ready waits up to 30 seconds and prints the combined status.

With --non-interactive and without --plan, one setup command submits directly. For a source that needs an API key, add --api-key-env NAME so the key is read from an environment variable rather than typed. All options are in the CLI reference.

Readiness states

status reports the first state that applies, in this order.

The dashboard and the terminal UI show the same statuses with the labels in Review and lifecycle.

Keep it running

--start and numes provider profile start run the endpoint in the background. It keeps running after you close the terminal, but not after a reboot. For a permanent setup, install it as a per-user service (a systemd user service on Linux, a LaunchAgent on macOS). Stop the background process first, because only one process can own a profile:

numes provider profile stop --profile ollama-localnumes provider service install --profile ollama-localnumes provider profile start --profile ollama-local

The service starts automatically; add --manual-start to install it without automatic startup. Once it's installed, profile start and profile stop control it. numes provider service status and numes provider service remove (both with --profile) show and remove it.

• Linux: user services start at boot and survive logout only with lingering enabled for your user. The command enables it when it can, and otherwise prints the command for an administrator, such as sudo loginctl enable-linger <user>.

• macOS: the service starts when you sign in.

• API keys: a service doesn't inherit your shell environment. If the profile reads a key from an environment variable (env://), service install copies its value from your current shell into the profile's owner-only secret store, and refuses with the variable's name when it isn't set. The desktop app's Background service panel asks for the same values.

Installing a service also installs the updater that keeps the agent current; see Automatic updates. On stop, the endpoint stops accepting new requests and lets running ones finish before it closes. The desktop app can install and manage the same services, including on remote machines over SSH.

Next steps