LLM API

Use the same Lime portal to call cloud models

OpenAI / Anthropic compatible portal for developers. The client only needs to access the Lime Base URL and Lime API Key, and the model directory, package permissions, quota reservation, and usage settlement are all handled by Lime.

Base URL

https://llm.limeai.run

Authentication

API Key + X-Lime-Tenant-ID

Model source

GET /v1/models returns the directory visible to the current tenant

Public network entrance

Only access limecore gateway-svc

Quick start

List the model first, then initiate a streaming request

Do not hardcode model names. First read the available models from the current tenant's model directory, and then fill in the model id into the compatible protocol request.

See OpenAI examples See Anthropic example

Get model directory

curl "https://llm.limeai.run/v1/models" \
  -H "Authorization: Bearer $LIME_API_KEY" \
  -H "X-Lime-Tenant-ID: $LIME_TENANT_ID"

OpenAI-compatible

Chat Completions

OpenAI SDKs or compatible HTTP clients can point the Base URL to Lime. Chat, Responses, Models, and Embeddings are part of the first-phase main path.

curl -N "https://llm.limeai.run/v1/chat/completions" \
  -H "Authorization: Bearer $LIME_API_KEY" \
  -H "X-Lime-Tenant-ID: $LIME_TENANT_ID" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "<model-id-from-/v1/models>",
    "messages": [
      { "role": "user", "content": "Hello, Lime" }
    ],
    "stream": true
  }'

Anthropic-compatible

Messages

Anthropic-compatible requests can use x-api-key, the gateway will retain the necessary protocol headers and replace them internally with real upstream authentication.

curl -N "https://llm.limeai.run/v1/messages" \
  -H "x-api-key: $LIME_API_KEY" \
  -H "X-Lime-Tenant-ID: $LIME_TENANT_ID" \
  -H "anthropic-version: 2023-06-01" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "<model-id-from-/v1/models>",
    "max_tokens": 1024,
    "messages": [
      { "role": "user", "content": "Explain the Lime LLM API in three sentences" }
    ]
  }'

unified entrance

OpenAI SDKs, Anthropic SDKs, and regular HTTP clients all enter through the same Lime gateway, so clients do not need to know about the provider layer behind it.

Business closed loop

API keys, tenants, packages, points, quota reservations, and usage submissions are all completed on the Lime side without exposing the upstream key.

streaming proxy

The main paths of Chat, Responses and Messages support JSON and SSE, and the response body is transparently transmitted according to the original protocol as much as possible.

Private online travel

new-api, sub2api and direct upstream are only internal routing candidates and are not used as the developer public network Base URL.

First phase main path

protocol	Endpoint	level	illustrate
Models	`GET /v1/models`	P0	Return to Lime visible model directory
Chat Completions	`POST /v1/chat/completions`	P0	OpenAI-compatible main path
Responses	`POST /v1/responses`	P0	Recommended main path for new projects
Anthropic Messages	`POST /v1/messages`	P0	Anthropic-compatible main path
Embeddings	`POST /v1/embeddings`	P0	Non-streaming input side billing

Boundaries should be clear

Developers only connect to Lime gateway

llm.limeai.run Is the current public network entrance. Clients should not access new-api, sub2api, or other provisioning layers directly.

The model directory is subject to the tenant

Different plans, tenants, and brand configurations will see different models. Actual availability is based on /v1/models Subject to return.

Long-tail protocols are open on demand

Files, Batches, Fine-tuning, and similar capabilities are not promised on the first-phase website. See the protocol coverage matrix for scheduling.

In-depth information

The official website only lists access entrances and stable boundaries. Protocol details, deployment topologies, and governance classifications continue to have the repository documentation as the source of truth.

Return to developer portal