Skip to content
Live 1 providers 8 models routable 1 keys

Grav AI Platform

One OpenAI-compatible endpoint in front of all your upstream providers. Multiple API keys per provider, load balanced and failed over automatically. Point it at any OpenAI-compatible URL: the models are detected from the endpoint itself and stored in the database — no model list in the code.

  • Open admin
  • Tool hosting
  • Try the API
    •

    Fleet health

    live
    Healthy keys 1 / 1
    Success ratio 96.7%
    90 requests · 3 errors
    Providers
    1
    1 enabled
    Models
    8
    routable right now
    Active keys
    1
    healthy: 1
    Requests
    90
    3 errors 96.7% ok

    Load spread

    Requests per provider; errors are drawn in red.

    90 total
    minimax
    ProviderStrategyModelsKeysHealthyRequestsErrors
    minimax round_robin 8 1 1 90 3

    Load spread

    Requests handled per provider, errors in red.

    90 total
    minimax

    Recent activity

    minimax minimax served 90 requests
    1/1 keys healthy · 8 models · 3 errors
    firecrawl-scrape tool firecrawl-scrape ready
    Scrape a single URL and return its content as markdown. A key is optional (keyless tier). · 0 calls
    firecrawl-search tool firecrawl-search ready
    Search the web and return the top results with their content. · 0 calls
    http-get tool http-get ready
    Fetch any URL over GET and return the raw body as text. · 0 calls
    pwn tool pwn ready
    no description · 0 calls

    Call it like OpenAI

    Point any OpenAI SDK at /proxy/ai/v1 and swap the base URL.

    curl https://ai.grav.cn/proxy/ai/v1/chat/completions \
  -H "Authorization: Bearer $PROXY_AUTH_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"",
       "messages":[{"role":"user","content":"hi"}]}'
    /proxy/ai/v1/models stream: true

    What you get

    LB 5 strategies — round robin, random, weighted, least used, failover
    HA 429/5xx/401 automatically retries on the next key
    AUTO Models are read from the upstream /v1/models
    UI Everything is editable from the browser, no restart needed
    TOOLS 3 of 4 hosted tools a model may call

    Try it

    Sends a real request through the proxy and shows the answer.

    /proxy/ai/v1/chat/completions

    • Live providers

    read-only

    minimax

    1/1 keys healthy round_robin
    MiniMax-M2 MiniMax-M2.1 MiniMax-M2.1-highspeed MiniMax-M2.5 MiniMax-M2.5-highspeed MiniMax-M2.7 MiniMax-M2.7-highspeed MiniMax-M3