Foundry
Start building

SDK · Adapters

Plug Foundry into your existing stack in three lines.

Foundry ships three first-class adapters. Each one is a thin translation layer over the same OpenAI-compatible HTTP proxy — so the inference path, receipts, and on-chain revenue routing are identical regardless of which adapter you choose.

Vercel AI SDK

Implements the LanguageModelV1 interface. Works with generateText, streamText, and generateObject.

vercel-ai.tsts
import { generateText } from "ai";
import { foundry } from "@foundryprotocol/sdk/adapters/vercel-ai";

const { text } = await generateText({
  model: foundry("ingot:0x8e2af4a…"),
  prompt: "Translate to Konkani: hello, how are you?",
});

console.log(text);

Streaming

ts
import { streamText } from "ai";
import { foundry } from "@foundryprotocol/sdk/adapters/vercel-ai";

const result = await streamText({
  model: foundry("ingot:0x8e2af4a…"),
  prompt: "Stream a haiku about co-owned models.",
});

for await (const chunk of result.textStream) {
  process.stdout.write(chunk);
}

LangChain

Implements a BaseChatModel-compatible class. Plugs into LCEL chains, agents, and RAG pipelines.

langchain.tsts
import { FoundryChat } from "@foundryprotocol/sdk/adapters/langchain";
import { HumanMessage, SystemMessage } from "@langchain/core/messages";

const llm = new FoundryChat({
  ingotId: "ingot:0x8e2af4a…",
  temperature: 0.6,
});

const res = await llm.invoke([
  new SystemMessage("You are a Konkani translation assistant."),
  new HumanMessage("Translate: where is the train station?"),
]);

console.log(res.content);
console.log(res.additional_kwargs.foundry.receipt);

LangChain is an optional peer dependency

We declare @langchain/core as an optional peer in package.json. If you don't install it, only the adapter module is unavailable — the rest of the SDK works identically.

OpenAI-compatible HTTP

Any tool that speaks the OpenAI API can call a Foundry Ingot. Point the base URL at api.foundryprotocol.xyz/v1 and pass the Ingot ID as x-foundry-ingot-id.

curlbash
curl https://api.foundryprotocol.xyz/v1/chat/completions \
  -H "content-type: application/json" \
  -H "x-foundry-ingot-id: 0x8e2af4a000000000000000000000000000000001" \
  -d '{
    "messages": [
      { "role": "user", "content": "Translate to Konkani: hello" }
    ],
    "stream": false
  }'

OpenAI's own SDK works out of the box:

openai-sdk.tsts
import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://api.foundryprotocol.xyz/v1",
  apiKey:  "not-required-but-the-sdk-insists",
  defaultHeaders: {
    "x-foundry-ingot-id": "0x8e2af4a000000000000000000000000000000001",
  },
});

const res = await client.chat.completions.create({
  model: "ingot:0x8e2af4a…",
  messages: [{ role: "user", content: "Hello" }],
});
console.log(res.choices[0].message.content);

Streaming

All three adapters stream tokens via Server-Sent Events in the OpenAI delta format. The final frame includes a foundryblock with the inference + revenue tx hashes.

final SSE frametext
data: {
  "id": "chatcmpl-foundry-…",
  "object": "chat.completion.chunk",
  "choices": [{ "index": 0, "delta": {}, "finish_reason": "stop" }],
  "foundry": {
    "ingotId":        "0x8e2af4a…",
    "inferenceTxHash":"0x4a7c…",
    "revenueTxHash":  "0x6f12…"
  }
}

data: [DONE]

Headers & receipts

request headerstext
x-foundry-ingot-id   0x…           required: which Ingot to call
authorization        Bearer …       optional: integrator API key for rate-limit
content-type         application/json
response headerstext
x-foundry-ingot-id   0x…   echo
x-foundry-stub       1     present on stub responses (Sprint 2/3); absent in prod

Next →

Foundry CLI