Your model, available wherever Spike is.

Before you begin

You need three things:

• A free Spike account, signed in on the Runtime host and on every client that should use it.
• LM Studio, Ollama or another OpenAI-compatible server running on the host machine.
• At least one model loaded and visible from the provider’s /v1/models endpoint.

Host from Spike for Mac

This is the simplest path for a Mac that already runs LM Studio, Ollama or another compatible provider.

Host from Spike Agent CLI

Use the CLI path for Linux, Windows, a headless server, or any machine where a foreground Runtime process is a better fit.

1. Install or update Spike

spike update
spike --version

You need Spike Agent CLI 2.0.17 or newer.

2. Sign into your Spike account

spike login
spike login status

3. Choose the local provider

Edit ~/.spike/config.toml and add one of these lines:

spike_runtime_provider = "lmstudio"

spike_runtime_provider = "ollama"

spike_runtime_provider = "home-model"

[model_providers.home-model]
name = "Home model"
base_url = "http://127.0.0.1:8000/v1"
wire_api = "chat"
env_key = "HOME_MODEL_API_KEY"

4. Start the Runtime

spike runtime serve --label "Home Runtime"

The Runtime runs in the foreground. Keep the process alive; press Ctrl-C when you intentionally want it to go offline. A long-running host can later be placed under launchd or systemd.

Connect your other clients

1. Sign into Spike Web, iPhone, Mac or CLI with the same Spike account.
2. Open model settings and choose My connected Runtime or the Runtime’s label.
3. Select one of the models advertised by that Runtime.
4. Send a short test prompt. Your host should show activity while the answer streams back.

If the Runtime appears as offline, first check that the Mac app or spike runtime serve process is still running.

How Runtime protects the local provider

The Spike account is a control plane, not the inference engine. It authenticates your devices, discovers the Runtime and relays a bounded model request to the host. The host calls only its configured private provider.

• The provider URL and provider key stay on the Runtime host.
• The Runtime makes an outbound connection; clients never need the host’s IP address.
• Runtime credentials are device-scoped and cannot access account or billing operations.
• Inference response bytes are relayed for the live request rather than becoming provider configuration in account storage.

Your own model and hardware perform inference. Spike Cloud entitlement is relevant only when you choose a Spike-hosted model.

Troubleshooting

“Runtime hosting is disabled”

The CLI did not find spike_runtime_provider in ~/.spike/config.toml, or the named provider does not exist. Add one of the configurations above and restart the command.

The Runtime is visible but offline

Keep Spike for Mac open, or restart spike runtime serve. Confirm every device uses the same Spike account.

Provider status is unavailable

Start the local model server, load a model, and check its URL. For LM Studio, the common base URL is http://127.0.0.1:1234/v1.

I receive 429

The request is reaching a rate-limited hosted path instead of your Runtime. Re-select the connected Runtime and confirm it is online. In the CLI, run spike login status to rule out an expired or missing account session.

I receive 401

Sign in again on that client. A Runtime provider key and a Spike account login are separate credentials and should never be substituted for each other.