@telaro/ai-gateway-adapter

One adapter import wraps the Telaro MCP toolset as an OpenAI-shaped function-calling tool. Drop the descriptor into any AI gateway that accepts the OpenAI tool shape (OpenRouter, Cloudflare AI Gateway, Vercel AI Gateway, direct OpenAI) and every model served through that gateway can call Telaro.

pnpm add @telaro/ai-gateway-adapter

Zero runtime deps. The adapter is a pure-JS translator. Network calls happen only when the model picks a tool and you run it.

Quickstart (OpenAI)

import OpenAI from "openai";
import { telaroGatewayTool, runTelaroTool } from "@telaro/ai-gateway-adapter";

const openai = new OpenAI();

const completion = await openai.chat.completions.create({
  model: "gpt-4o-mini",
  messages: [
    { role: "user", content: "find me the top three bonded kamino agents" },
  ],
  tools: [telaroGatewayTool({ apiKey: process.env.TELARO_API_KEY })],
});

const toolCall = completion.choices[0].message.tool_calls?.[0];
if (toolCall) {
  const result = await runTelaroTool(toolCall, {
    apiKey: process.env.TELARO_API_KEY,
  });
  console.log(result);
}

OpenRouter

OpenRouter accepts the same OpenAI tool shape. Pass the descriptor through unchanged:

import { telaroGatewayTool } from "@telaro/ai-gateway-adapter";

const r = await fetch("https://openrouter.ai/api/v1/chat/completions", {
  method: "POST",
  headers: {
    Authorization: `Bearer ${process.env.OPENROUTER_API_KEY}`,
    "Content-Type": "application/json",
  },
  body: JSON.stringify({
    model: "anthropic/claude-sonnet-4.6",
    messages,
    tools: [telaroGatewayTool()],
  }),
});

Cloudflare AI Gateway

Same shape. Tools forward through the AI Gateway's tool-call plumbing.

const r = await fetch(
  `https://gateway.ai.cloudflare.com/v1/${ACCOUNT}/${GATEWAY}/openai/chat/completions`,
  { method: "POST", headers, body: JSON.stringify({ model, messages, tools: [telaroGatewayTool()] }) },
);

Vercel AI SDK

Use the v6 tools field. Convert the descriptor to the AI SDK shape with the helpers already shipped in @vercel/ai:

import { generateText } from "ai";
import { telaroGatewayTool } from "@telaro/ai-gateway-adapter";

await generateText({
  model: "anthropic/claude-sonnet-4.6", // AI Gateway model string
  messages,
  tools: { telaro: telaroGatewayTool() as any },
});

Two shapes

telaroGatewayTool() returns a single umbrella tool the model uses by setting action. Most gateways prefer this because it eats one slot in the tool budget rather than six.

telaroGatewayToolsExpanded() returns six explicit tools. Use this if your gateway prefers tool-name dispatch (and you have room).

Both work with the same runTelaroTool runner.

Tools

| Tool | What it does | |---|---| | agent_lookup | Full trust profile of a bonded agent. | | agent_list | List bonded agents, filtered by framework. | | credit_balance | Read the connected wallet's prepaid balance. | | credit_charge | Debit the balance for a tool call. | | disputes_open_list | Open disputes the wallet can act on. | | dispute_open | File a claim with a 5% deposit. |

Auth

| Header | When to send | Example | |---|---|---| | x-api-key: <key> | Read endpoints with rate limit | apiKey: process.env.TELARO_API_KEY | | Authorization: Bearer <token> | Write endpoints (credit_charge, dispute_open) | bearerToken: oauthToken |

The bearer token is the OAuth grant from the end user. Hot-key rotation via AgentKeyLink is supported on the wallet side; the adapter never sees the wallet's private key.

License

MIT.