Developers

Build oninterest.

One OpenAI-compatible API for every model, paid from your credit. From your code, from your agent over MCP, or from an agent's own wallet.

Works with LangChainAI SDKLlamaIndexCursorClaude Code
⌘

OpenAI-compatible

Streaming, tools, JSON mode and images pass straight through. LangChain, the Vercel AI SDK and LlamaIndex work as they are.

⇄

Over MCP

Claude, Cursor and other agents call any model with your credit: a second opinion, a cheaper hand-off, a model they lack.

⌥

Keys with limits

One key per app, each with its own spending limit, allowed models and a pause switch.

Quick start

Four steps to
your first request.

  1. 1

    Connect a wallet

    Open the app and connect your Solana wallet.

  2. 2

    Get credit

    Deposit into a vault to earn it, or mint it with USDC.

  3. 3

    Create a key

    Under Developers. It is shown once; give each app its own.

  4. 4

    Point your SDK

    Set the base URL and the key. Everything else stays the same.

Agents

Agents that
pay their own way.

An agent signs in with its own Solana wallet, holds its own credit, and tops itself up from its own vault's interest. No borrowed key, no card on file.

research-bot

MCP server

claude mcp add --transport http \
  stipendo $STIPENDO_BASE_URL/mcp

Tools: ask_model, list_models, get_balance.

Billing

Every request is paid from your credit: the tokens the model used, at the provider's price plus 5%. Every charge shows with tokens, cost, time and key.

Routing only

Keep your own OpenAI or Anthropic account. Send the request, get back the model to call. A flat $0.0002 per decision.

Response headers

x-stipendo-model, x-stipendo-tier and x-stipendo-cost tell you what answered and what it took.

Let your stock
pay the bill.

Launch app ↗