路墨 LumoswitchAI+ application integration
Connect several model services through one compatible endpoint and keep a stable calling surface for the application.
An AI+ inference management platform for developers and enterprises in China
Connect your models.Ship reliable AI apps.
Use one OpenAI-compatible endpoint to manage DeepSeek, Qwen, Doubao, Kimi, and other configured model services. Give AI agents, enterprise knowledge assistants, and smart-factory applications one stable entry point, with routing, access keys, limits, usage, and billing in one console.
OpenAI-compatible API Bring your own key or use platform models Usage attributed per request
One entry point, many AI models
Connect your existing AI tools to Lumoswitch, and manage model choices and switching here.
Your AI toolsLumoswitchModels you choose
Your chosen order
Use your selected model first. If it is unavailable, try backup models according to your settings.
DeepSeekPreferred
QwenBackup
DoubaoBackup
Smart factories and business agents
Configure available model resources consistently for inspection, Q&A, and workflow-collaboration agents.
Model capabilities for teams
Organize model configurations and access keys by project; configurations define model scope and call rate so boundaries stay clear.
New-user promotional credits
Promotional credits are deducted first, while model usage and billing details remain traceable in the console.
Compatible with leading China-based model services and compatible providersAvailable models depend on the current deployment and resources published in the console
DeepSeekQwenDoubaoKimi
Capabilities
One control plane, from model access to usage accounting
Clients only make requests; models, routing, and access boundaries remain adjustable in Lumoswitch.
Unified multi-model access
Manage your own keys, OpenAI-compatible providers, and platform-published models in one place instead of maintaining integration code for every service.
- Unified model catalog
- Encrypted credential storage
- Your own keys alongside platform models
Configurable runtime routing
Fixed, failover, automatic selection, least-busy, and lowest-latency policies run only among the model candidates you select.
- Controlled candidate order
- Independent downstream model IDs
- Visible configuration sync status
Keys and access boundaries
Create independent access keys for model configurations, where available models and call rate are constrained; rotate or revoke each key independently.
- Isolated model configurations
- Configuration-level rate limits
- Key rotation and revocation
Usage, balances, and billing
Attribute model calls to users and keys, separate promotional credits from topped-up balances, and retain token, cost, and ledger records.
- Promotional credits used first
- Multi-dimensional usage metering
- Traceable billing details
Quick start
From an existing model key to one endpoint in three steps
No business logic rewrite is needed. After creating a configuration, put the shared endpoint, access key, and model ID into your existing client.
- 01Connect model resources
Add keys for existing model services, or select platform models already published by an administrator.
- 02Configure routing and model IDs
Choose model candidates and a scheduling method, then define the downstream model ID that clients use.
- 03Create an access key
Copy the Base URL, key, and model ID into an existing AI application or agent.
import OpenAI from "openai";
const client = new OpenAI({
apiKey: process.env.LUMOSWITCH_API_KEY,
baseURL: "https://api.lumoswitch.com/v1",
});
await client.chat.completions.create({
model: "production-chat",
messages: [{ role: "user", content: "Hello" }],
});Request record examplereq_7A42F1
Request authorizedAccess key, model configuration, and account status passed validation
429
Primary model rate-limitedContinuing in the saved candidate order
200
Standby model completedThe client received a normal result
Usage attributedTokens, model, key, and cost entered billing
Reliability and observability
Switch models without changing clients; every request is recorded
The client always calls the same model ID. When rate limits or service failures occur, runtime behavior follows only your configured rules and writes the resolved model, status, latency, and usage to the request record.
- Clear scheduling boundariesCandidates, order, and policy are set by the configuration and never exceed its authorized scope.
- Layered balance recordsPromotional credits are used first; topped-up balances and refundable amounts remain separate.
- Faster issue diagnosisA clear trail is retained from key and configuration to model and request record.
Use cases
From personal development to platform servicesuse one model entry point
From a single agent to user-facing model capabilities, manage clear key, routing, and billing boundaries.
AI agents and developer tools
Let Codex, OpenClaw, OpenCode, or internal agents use one endpoint, while model changes are handled in the console.
A model-service entry point for teams
Store upstream credentials centrally, give projects their own model configurations and access keys, and keep balances and billing at the account level.
Resilient multi-provider calls
Map one business model to multiple candidate resources and, when configured, switch in order to reduce the impact of a single provider failure.
FAQ
Questions you may have before integrating
These answers describe product boundaries. Actual models, prices, credits, and interfaces depend on the current deployment and console.
What is the relationship between 路墨 and Lumoswitch?
路墨 Lumoswitch is the Chinese brand name for Lumoswitch and refers to the same product capabilities, not another AI agent or a foundation model. It connects existing AI tools to configured model services through one endpoint and centralizes routing, access keys, limits, usage, and billing.
Read the full answer: What is the relationship between 路墨 and Lumoswitch?Do I need to rewrite an existing OpenAI SDK integration?
Usually not. Replace the Base URL, API key, and model ID with the values supplied by Lumoswitch and continue using an OpenAI-compatible client. Supported formats depend on your configuration and documentation.
Read the full answer: Do I need to rewrite an existing OpenAI SDK integration?How can Codex use our team model endpoint?
Codex workflows using an OpenAI-compatible integration can be configured with the team Lumoswitch Base URL, access key, and public model ID. Exact fields depend on the client version and deployment; consult that client's documentation.
Read the full answer: How can Codex use our team model endpoint?What happens if the primary model is unavailable?
Only a configured failover policy attempts later candidates at runtime. Fixed routing never switches automatically; other scheduling modes follow the saved candidate scope and rules.
Read the full answer: What happens if the primary model is unavailable?How are API keys, usage, billing, and request logs managed?
Create separate access keys for model configurations, use those configurations to limit available models and rate, and manage balances and billing at the account level. Keys can be rotated or revoked independently. Actual usage, active credits, and data boundaries are shown in the console and applicable policies.
Read the full answer: How are API keys, usage, billing, and request logs managed?Lumoswitch China edition
Start building your unified model entry point
Connect one model resource first, then gradually move routing, keys, limits, and usage management into one console.