About LYRouter
LYRouter is a trusted model service and token governance platform.
- Trusted model services: applications and coding tools call models from many vendors through one address,
https://api.lyrouter.com, and one LYRouter API key. Every call goes through the same authentication, routing and guardrails, and leaves a complete call log. - Token governance: token usage on platform models is billed in Credits, and the usage of every call is attributed to workspaces, members, departments and API keys; you can set budgets and quotas to control usage, and lower costs with routing policies and prompt caching.
You can use models provided by the platform and pay by usage, or connect your own API keys from providers (BYOK).
What LYRouter does
Trusted model services
| Capability | Description | Details |
|---|---|---|
| Unified access | Supports three protocols, OpenAI Chat Completions, Responses and Anthropic Messages, plus embeddings, rerank, image, video and audio endpoints. Existing SDKs and tools only need a new address and key | Base URLs and authentication |
| Two model sources | Platform models deduct Credits by usage; BYOK uses your own provider account, and the provider bills you directly | Platform models or BYOK |
| Routing and failover | The channels of a model are sorted by price, latency or throughput, and the next channel takes over automatically on failure. You can also pin a set of models and parameters with a virtual model, or use lyrouter/auto to pick a model by task automatically |
Channels and routing |
| Teams and security | Organization, department and member management, workspace authorization, and guardrails such as model access control and content safety detection | Teams, members and roles |
Token governance
| Capability | Description | Details |
|---|---|---|
| Costs and limits | Everything is billed in Credits; you can set budgets on API keys and quotas on members, departments or workspaces | Billing and Credits |
| Usage attribution and observability | Every call has a request log that shows the actual model, tokens, cost and failure reason; usage analytics summarize by model, API key and member; the service status page publicly shows the health of each upstream line | Troubleshooting with logs |
| Cost optimization | Choose channels by price and raise the prompt cache hit rate to pay less for repeated input | Reducing costs and improving cache reuse |
Editions
| Edition | Suited to | Main features |
|---|---|---|
| Personal | Individual developers | Free, one default workspace, BYOK available |
| Team | Team collaboration | Per-seat subscription. You can invite members, set up departments, create several workspaces, and use guardrails such as quotas, model access control and client access control. Creating one requires the Owner to complete personal identity verification |
| Enterprise | Organizations with compliance requirements | Adds sensitive data, prompt injection and intent security detection to Team, and keeps logs and security events longer. Creating one requires the Owner to complete personal identity verification first, then enterprise verification |
For each edition's exact allowances, such as seats, number of workspaces and log retention, go by the plan cards on the console's Subscription & Credits page; see Billing and Credits.
Where to start
| You want to | Read |
|---|---|
| Make your first call from code | Your first model call |
| Use LYRouter in tools such as Claude Code and Codex | Tools and compatibility |
| Create a workspace and API keys | Workspaces and API keys |
| Understand concepts such as accounts, workspaces and model sources | Core concepts |
| Connect your own provider account | Connect a provider |
| Invite members and manage permissions | Teams, members and roles |
| Find out why a call failed | Troubleshooting with logs |
| Look up API parameters and what errors mean | Base URLs and authentication, Errors and request identifiers |
Models differ in the parameters, context length and input types they support. Check them in the workspace's Model marketplace before you integrate; see Choosing a model.