Skip to content

linkapi.md

Latest commit

 

History

History
162 lines (129 loc) · 5.47 KB

linkapi.md

File metadata and controls

162 lines (129 loc) · 5.47 KB
Raw

LinkAPI chat and image generation

MultiLLM-Proxy preserves LinkAPI's native OpenAI, Anthropic, and Gemini protocols. The proxy does not translate image payloads or responses, so use the endpoint listed for the selected model in LinkAPI's live catalog.

Official references:

The deployed default is LinkAPI's HK direct origin, https://hk.linkapi.ai. Set LINKAPI_BASE_URL to another allowlisted official origin only when regional measurements justify it; both the direct Worker path and controlled Flask path use the same configured origin.

Configuration

Set the upstream LinkAPI credential as LINKAPI_KEY. LINKAPI_API_KEY remains an alias, but LINKAPI_KEY is preferred.

Direct /linkapi/* requests use ADMIN_API_KEY as the caller credential. The Worker removes it and authenticates upstream with LINKAPI_KEY.

export PROXY_BASE_URL="https://your-worker.example"
export ADMIN_API_KEY="your-proxy-key"

Discover current models

LinkAPI's catalog is dynamic, so query it instead of hard-coding model IDs:

curl "$PROXY_BASE_URL/linkapi/v1/models" \
  -H "Authorization: Bearer $ADMIN_API_KEY"

The direct route returns LinkAPI's response unchanged.

Generate with GPT Image

gpt-image-2-c exposes LinkAPI's OpenAI Images generation and edit contracts.

Use the unified route when you need the proxy's normal request-size checks, rate limits, accounting, and metrics:

curl -X POST "$PROXY_BASE_URL/v1/images/generations" \
  -H "Authorization: Bearer $ADMIN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "linkapi:gpt-image-2-c",
    "prompt": "A navy command room at dusk, cinematic lighting",
    "size": "1024x1024",
    "quality": "standard",
    "moderation": "low",
    "style": "vivid",
    "n": 1,
    "response_format": "url"
  }'

Use the raw Worker route for the lowest proxy overhead:

curl -X POST "$PROXY_BASE_URL/linkapi/v1/images/generations" \
  -H "Authorization: Bearer $ADMIN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-image-2-c",
    "prompt": "A navy command room at dusk, cinematic lighting",
    "size": "1024x1024",
    "quality": "standard",
    "moderation": "low",
    "style": "vivid",
    "n": 1,
    "response_format": "url"
  }'

LinkAPI's live model detail shows response_format with url as the default. The proxy preserves the selected upstream-supported format and all safe response metadata.

For GPT Image generation, moderation accepts low or auto; MultiLLM inserts low when the field is omitted. This default is identical on the unified Container route and the direct Worker route.

Image editing is available on the direct multipart route:

curl -X POST "$PROXY_BASE_URL/linkapi/v1/images/edits" \
  -H "Authorization: Bearer $ADMIN_API_KEY" \
  -F "model=gpt-image-2-c" \
  -F "prompt=Add soft cinematic lighting" \
  -F "image=@input.png"

The unified route currently handles JSON image generation only. Use the direct route for multipart edits.

Generate with Gemini Flash Image

Gemini image models expose LinkAPI's native Gemini generateContent contract. Native Gemini is preferred because it preserves image parts exactly:

curl -X POST \
  "$PROXY_BASE_URL/linkapi/v1beta/models/gemini-2.5-flash-image:generateContent" \
  -H "x-goog-api-key: $ADMIN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "contents": [{
      "parts": [{
        "text": "Create a watercolor lighthouse at dawn"
      }]
    }],
    "generationConfig": {
      "responseModalities": ["TEXT", "IMAGE"]
    }
  }'

The response remains Gemini-native. Read image data from the returned content parts; MultiLLM-Proxy does not convert those parts into OpenAI data[] objects.

Current image-model snapshot

This snapshot was verified against LinkAPI Standard pricing on July 23, 2026. The live catalog, prices, group ratios, and per-model limits can change.

Model Pricing shown
gemini-2.5-flash-image ¥0.08/request
gemini-3-pro-image ¥3.20 input and ¥192 output per 1M tokens
gemini-3-pro-image-preview ¥0.35/request
gemini-3.1-flash-image ¥0.80 input and ¥96 output per 1M tokens
gemini-3.1-flash-image-preview ¥0.15/request
gemini-3.1-flash-lite-image ¥0.08/request
gpt-image-2 ¥28 input, ¥105 output, and ¥7 cached input per 1M tokens
gpt-image-2-c ¥0.10/request

For low-cost volume, start by testing gemini-2.5-flash-image or gemini-3.1-flash-lite-image. Use gpt-image-2 when its output quality or OpenAI Images contract is more important than the lowest unit price. Confirm quality, daily limits, and the endpoint shown in each model's live detail page before committing production traffic.

Operational behavior

  • /linkapi/* is a raw Worker fast path on Cloudflare and does not wake the Flask Container.
  • The raw path intentionally bypasses Flask dashboard-user authentication, application request-size checks, RPM/TPM/daily accounting, and metrics.
  • /v1/images/generations uses the controlled Flask path and requires a provider:model ID such as linkapi:gpt-image-2-c.
  • Generation requests are single-attempt. The proxy does not retry paid image calls or promise idempotency.
  • Do not assume a subscription can sustain 10,000 image requests per day. Check the selected model's current RPM and RPD limits in LinkAPI before load testing.