Installations

SuperGPT for Mac is being qualified.

The Apple-Silicon source contour is ready for integration, but public DMG and ticket issuance are intentionally paused until the exact Gateway catalog, activation envelope, live fleet canary and final artifact all bind to one release receipt.

Open App → API

SuperGPT DMG delivery is currently paused

The dashboard will verify the DMG receipt and issue a one-use ticket only after every release gate is green. Until then, no public installer artifact is advertised.

Planned macOS Apple Silicon release

1The current local-mirror release is an internal source candidate; public DMG delivery remains paused.
2The source contract admits subscription text and funded API-wallet text; Dictation is subscription-only and remains disabled pending Product DB/Gateway entitlement rollout and immutable clean-VM evidence.
3After qualification, the dashboard will expose one checksum-verified, user-approved ad-hoc helper DMG; normal setup uses one-time device pairing and never reveals the product key.
4Windows and Intel Mac delivery remain excluded until they pass independent release gates.

Exact provider values

These are the fallback fields for any OpenAI-compatible IDE form.

Provider ID

luma-cloud

Display name

Luma Cloud

Provider type

OpenAI-compatible

Base URL

https://api.lumaos.cloud/v1

Models

Read authenticated GET https://api.lumaos.cloud/v1/models and use one exact returned id

Base-model default effort

medium, only when the caller sends no explicit effort

Family context

GPT-5.5 272K; GPT-5.6 372K

Requested output cap

128000

OpenCode timeout

600000

OpenCode chunkTimeout

300000

Client matrix

One provider contract for every coding surface.

Each client gets the smallest correct setup path: a secret-free profile where useful, or the Base URL and API key in its OpenAI-compatible form.

Guide

OpenCode

Download secret-free config

Create an API key in App → API, keep it in LUMA_CLOUD_API_KEY, then download a provider skeleton with no raw secret or static model list.

Open guide

Release candidate

SuperCodex

Planned user-approved helper DMG

Apple Silicon source is prepared, but public delivery stays paused until Product DB/Gateway entitlement and immutable clean-Mac qualification pass.

Open guide

Guide

OpenClaw

Use the OpenClaw guide

OpenClaw uses the public Luma Cloud endpoint plus an explicitly created API key and canonical model id.

Open guide

Manual setup

Cursor

Custom OpenAI provider

Use the base URL, bearer key, and an exact lowercase model id returned by the authenticated /v1/models endpoint.

Open guide

Manual setup

Continue / VS Code

OpenAI-compatible model block

Set apiBase or baseURL to the Luma Cloud /v1 endpoint and use the API key created in App → API. The supported model list is generated from the shared catalog.

Open guide

Manual setup

Cline

OpenAI-compatible API provider

Use the /v1 endpoint and the API key created in your dashboard. Do not paste localhost or other internal URLs.

Open guide

Manual setup

Roo Code

OpenAI-compatible API provider

Use the same endpoint, key, and model ids as Cline. Verify native tool-call streaming before claiming full tool support.

Open guide

Manual setup

Zed

OpenAI-compatible settings

Add the Luma Cloud api_url and available model ids to the Zed assistant/provider settings.

Open guide

Provider UX

Why OpenRouter is one step, and how Luma Cloud gets there.

The goal is the same: provider id, endpoint, key, model. Manual OpenAI-compatible setup works now; native catalog listing is a separate provider-adoption track.

Why OpenRouter feels automatic

OpenRouter is already listed in OpenCode's provider flow, and many OpenRouter models are preloaded by OpenCode. The user adds a key; OpenCode already knows the provider and model catalog.

What custom providers require today

OpenCode's Other provider path stores a credential first, then expects provider metadata and configured models. The secret-free skeleton supplies metadata; exact model ids still come from the authenticated catalog.

What Luma Cloud now exposes

The public API serves authenticated /v1/models for SDKs and a secret-free OpenCode provider skeleton. The Custom provider form remains the simplest path today.

Fallback

Manual setup stays short when a client has only a form.

Use this block for Cursor, Continue, VS Code extensions, local SDK tests, and any client that asks for an OpenAI-compatible provider.

Copy-safe manual block

Always use the /v1 endpoint from your dashboard. Never paste localhost or other internal URLs into client settings.

Provider ID: luma-cloud
Display name: Luma Cloud
Provider type: OpenAI-compatible
Base URL: https://api.lumaos.cloud/v1
API key: explicitly created key from Luma Cloud App → API
Models: fetch the authenticated https://api.lumaos.cloud/v1/models response and use one exact returned id
OpenCode timeout: 600000
OpenCode chunkTimeout: 300000

What the optional provider skeleton contains

The downloaded JSON contains public provider settings and an environment reference, never a stored API secret.

provider.luma-cloud with @ai-sdk/openai-compatible

baseURL set to https://api.lumaos.cloud/v1

environment reference for the API key created in App → API

no static model list; fetch exact ids from authenticated GET /v1/models

limit.output set to 128000

safe request timeouts and no preselected default model

Other clients

No dedicated guide? Manual setup still works.

Any tool with an OpenAI-compatible provider form can connect today with the base URL, key, and model values from this page. The full universal path lives in the API setup guide.

JetBrains AI Assistant

Manual setup

Windsurf

Manual setup

AnythingLLM

Manual setup

LangChain / OpenAI SDKs

Manual setup

Troubleshooting

When the provider form still fails.

These checks separate bad local setup from a real backend issue.

Request body mismatch

Replace the old provider profile. Old OpenCode profiles often advertise output 1024 instead of 128000, or omit timeout/chunkTimeout, which can truncate a valid long reasoning request.

401 or invalid key

Create or replace the API key in App → API. Do not use the endpoint as the key.

Models not visible

Restart OpenCode after saving the provider and refresh the authenticated live catalog. Do not type a model omitted by GET /v1/models.

OpenCode opens empty after a bad config

Rename ~/.config/opencode/opencode.json to opencode.broken-lumacloud.json, restart OpenCode, then add the small provider block from the OpenCode guide.

Quota exhausted

The provider is connected. Open Billing or Analytics to inspect the account limit.

Wrong endpoint

Use the base URL ending in /v1. Do not append /chat/completions in provider settings.

Wrong model label

Use an exact lowercase id returned by GET /v1/models. Marketing labels like GPT-5.5X High are not copy-safe field values.

Backend contract

All public clients should enter through https://api.lumaos.cloud/v1.

OpenCode details API setup Provider adoption