A privacy-focused payment gateway that enables anonymous micropayments using e-cash notes (also known as Cashu notes) for accessing Large Language Models via the OpenAI API.

otrta Gateway is a privacy-focused payment gateway that enables anonymous micropayments using e-cash notes (also known as Cashu notes) for accessing Large Language Models via the OpenAI API. On the client side, it provides a Local Proxy that manages these notes using the Local Proxy Wallet. The Local Proxy modifies API requests to include the notes. On the server side, the 402 Server acts as a wrapper around the OpenAI API, handling note redemption, payment processing, and change generation with the help of the Server's Wallet.

• Anonymous AI Access: Access OpenAI's language models without revealing your identity
• Change Management: Returns unused funds through a Cashu note-based approach
• Fee Handling: Options to manage mint fees through change notes or private fee-free mints
• API Compatibility: Seamlessly works with standard OpenAI API endpoints

The diagram below illustrates the interaction between client and server components, each with their own wallet:

sequenceDiagram
    box Client Side
    participant CW as Local Proxy Wallet
    participant C as Local Proxy
    end
    box Server Side
    participant S as 402 Server
    participant SW as Servers Wallet
    end

    Note over CW,SW: Both Local Proxy and 402 Server have their own wallets

    C->>CW: 1. Prepare e-cash note for payment
    CW->>C: 2. Return the note
    C->>S: 3. Request + X-Cashu header
    S->>SW: 4. Process the payment from header
    SW->>S: 5. Extracted payment amount
    S->>S: 6. Execute requested operation
    SW->>S: 7. Calculate change (if overpaid)
    S-->>C: 8. Response + X-Cashu header (if applicable)
    C->>CW: 9. Store the change for future use (if received)

In this workflow:

1. The Local Proxy prepares a payment using the Local Proxy Wallet for an OpenAI API request
2. The Local Proxy Wallet provides a valid e-cash note
3. The Local Proxy sends the LLM request with the note in the X-Cashu header to the 402 Server
4. The 402 Server extracts the payment from the header and passes it to the Server's Wallet for processing
5. The Server's Wallet informs the 402 Server of the available amount
6. The 402 Server forwards the request to OpenAI and processes the AI model response
7. The Server's Wallet calculates change if the payment exceeds the actual usage
8. The 402 Server sends the OpenAI response with change in the X-Cashu header (only if overpaid)
9. The Local Proxy stores the change in the Local Proxy Wallet for future LLM requests (if change was provided)

This approach maintains privacy while efficiently handling micropayments for AI services. The 402 Server extracts the exact amount needed from the X-Cashu header for the specific OpenAI API call and returns any remainder via the X-Cashu header only when overpayment occurs.

The Bitcoin network's limitation of 1 satoshi as the smallest transaction unit creates challenges for micropayments in AI services. This constraint particularly affects high-volume, low-cost AI API calls where transaction amounts are often fractions of a satoshi.

The following flow chart visualizes how we intend to address the problem:

flowchart TD
    A[LLM API Request] --> B{Payment Amount?}
    B -->|< 1 sat| C[Problem: Can't send fractional sats]
    B -->|≥ 1 sat| D[Standard Payment]
    C --> E[User pays full sat]
    C --> F[Service waits for multiple requests]
    E --> G[Wasted value]
    F --> H[Complex service]

This flowchart illustrates the fundamental challenge with Bitcoin micropayments for AI services:

• When an OpenAI API call costs less than 1 satoshi, there's no native way to pay the exact amount
• Users must either overpay (wasting value) or the service must batch multiple LLM requests (more complexity)
• This inefficiency becomes significant for high-volume, low-cost AI API calls like embeddings or short completions

Wallet Gateway addresses the micropayment challenge for AI services through an innovative approach using millisatoshi (msat) precision Cashu mints. This method enables exact payments for API calls without rounding errors or ceiling effects, ensuring users pay precisely for what they consume.

The system uses HTTP headers for seamless payment integration: clients send ecash notes via the X-Cashu header, and servers return change (when overpaid) via the X-Cashu header.

Wallet Gateway addresses the micropayment challenge for AI services through this millisatoshi-based approach:

flowchart
    A[LLM Request + X-Cashu header] --> B[Gateway]
    B --> C{Process Payment}
    C --> D[Extract ecash from header]
    C --> E[Execute OpenAI API Call]
    C --> F[Calculate Exact Cost in msats]
    F --> G{Payment > Cost?}
    G -->|Yes| H[Return AI Response + X-Cashu header]
    G -->|No| I[Return AI Response only]

flowchart LR
    subgraph "Header-Based Payment Flow"
    I[X-Cashu] --> J[Exact msat Payment Processing]
    J --> K[Precise OpenAI API Billing]
    K --> L{Overpayment?}
    L -->|Yes| M[X-Cashu header]
    L -->|No| N[No change needed]
    end

This flowchart shows our solution:

• Clients send ecash notes through the X-Cashu HTTP header for seamless integration
• The gateway processes payments using millisatoshi precision for exact billing
• Payments are calculated to exact millisatoshi amounts based on actual API usage
• Change is returned via X-Cashu header only when the payment exceeds consumption
• This enables true micropayments for AI services without rounding errors or overpayment waste

There are two approaches to handling mint fees:

1. Change Cashu notes: Return unused funds as Cashu notes for future use with AI services
2. Private Fee-Free Mint: Operate a private mint without fees

• Gateway: API server handling Cashu note redemption and payment processing for OpenAI requests
• Wallet: Manages eCash (Cashu) notes and communication with mints
• Pay: Handles cost calculation and payment verification for AI model usage

1. Get initial Cashu note: Obtain a Cashu note from a Cashu mint
2. Make OpenAI API request: Send the Cashu note in the X-Cashu header with your LLM API request
3. Process response: Receive the OpenAI response and check for X-Cashu header
4. Save change (if any): Store any returned change Cashu note from the X-Cashu header
5. Use for next request: Use the change Cashu note in future X-Cashu headers for subsequent AI service requests

This approach allows for efficient micropayments for AI services. You only receive change when you overpay, preventing waste on small transactions.

Here's an example of how to make a request with ecash payment using curl:

curl -i -X POST https://ecash.otrta.me/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "X-Cashu: cashuAeyJ0b2tlbiI6W3sicHJvb2ZzIjpbeyJpZCI6IjAwOWExZjI5M2F..." \
  -d '{
    "model": "meta-llama/llama-4-scout-17b-16e-instruct",
    "messages": [
        {
            "role": "user",
            "content": "hello ecash"
        }
    ]
  }'

Example Response Headers:

...
X-Cashu: cashuAeyJ0b2tlbiI6W3sicHJvb2ZzIjpbeyJpZCI6IjAwOWExZjI5M2F...
...

{
  "choices": [
    {
      "message": {
        "role": "assistant",
        "content": "Hello! I see you're using ecash for payment. How can I help you today?"
      }
    }
  ]
}

The -i flag ensures you see the response headers, including any X-Cashu header with your change.

Contributions are welcome! Please feel free to submit a Pull Request.