Docs

The gateway reference and the staking rules, in one page.

Gateway

The gateway takes standard OpenAI chat requests. Set the base URL and send your key as a Bearer token.


        
Streaming works with "stream": true. Tool definitions, response formats and sampling parameters are forwarded unchanged to the provider that serves the model.

        
Checking gateway status.

      

      

        
Model list

        
GET /v1/models returns every model id in the OpenAI list format. Use the id exactly as returned, for example anthropic/claude-opus-5. Live prices for each id are on the pricing page and in GET /api/models.

      

      

        
Free providers

        
The gateway routes by the prefix of the model id. Ids with no listed prefix go to OpenRouter, where every id ending in :free costs nothing and shares a quota of about 50 requests a day per account. The other prefixes reach providers with their own free tiers, and each one is live once its key is connected.

        
    
          
  • nim/ NVIDIA NIM, free credits on signup, for example nim/meta/llama-3.3-70b-instruct
    • 
          
  • groq/ Groq, up to 14,400 requests a day on the small models, for example groq/llama-3.3-70b-versatile
    • 
          
  • cerebras/ Cerebras, one million tokens a day, for example cerebras/llama3.1-8b
    • 
          
  • gemini/ Google AI Studio, for example gemini/gemini-2.5-flash-lite
    • 
        

        
The connected upstreams are listed in GET /api/models under upstreams. A prefixed model whose upstream is not connected returns 503.

      

      

        
Errors

        
Errors use the OpenAI error shape: {"error": {"message", "type"}}.

        
    
          
  • 400 the body is missing model or messages
    • 
          
  • 401 the key is missing or not recognised
    • 
          
  • 503 the gateway is not accepting requests, or the upstream for that prefix is not connected
    • 
          
  • Any other status is passed through from the upstream provider
    • 
        

      

      

        
Staking

        
$SOCIETAL staking runs on Robinhood Chain in three steps.

        
    
          
  1. Stake $SOCIETAL 1:1 for sSOCIETAL. Unstaking has a seven day wait.
    2. 
          
  2. Lock sSOCIETAL to mint SCU. Redeeming burns the SCU and releases the locked sSOCIETAL.
    3. 
          
  3. Stake SCU. Staked SCU earns model credits every day at the rate published for that credit period, before it starts.
    4. 
        

        
Credits are a gateway balance. They pay for tokens and agent compute and cannot be withdrawn as tokens.

      

      

        
Verify on chain

        
The token contract address: issuance pending. Staking contract addresses are listed here at issuance. Before you stake, check that each address has contract code on Robinhood Chain and that the token symbol reads SOCIETAL.

        
Model credits and $SOCIETAL carry risk. Token prices move, smart contracts can fail, and credit rates change between periods. Nothing on this site is financial advice.