Framework adapters

@spliceloom/adapters gives the agent framework you already use Splice's tools: the read-only live-data tools (Robinhood Chain tokens, stock tokens, perps, DeFi, oracle prices, wallets, global markets) and the tools of your project's installed skills.

npm install @spliceloom/adapters      # Node.js 22.18+

The package has no dependency on any framework. You pass the framework's own tool factory, so the framework stays your dependency, at your version.

Collect the tools

import { spliceTools } from "@spliceloom/adapters";

const tools = await spliceTools({
  project: "./agent",              // installed skills come from this Splice project
  include: ["tokens_rank", "stock_quote", "defi_overview"], // optional allowlist
});

Each tool is a plain object: name, description, parameters (JSON Schema) and execute(args). Names are valid for every framework ([a-zA-Z0-9_-], at most 64 characters). execute never throws; a failure comes back as { status: "ERROR", message } for the model to read.

OpenAI Chat Completions

Works with OpenAI and any compatible endpoint (OpenRouter, local servers).

import { openAITools, runOpenAIToolCalls, spliceTools } from "@spliceloom/adapters";
import OpenAI from "openai";

const client = new OpenAI();
const tools = await spliceTools({ skills: false });
const messages: any[] = [{ role: "user", content: "What is the TVL of Robinhood Chain?" }];

for (;;) {
  const { choices } = await client.chat.completions.create({ model: "gpt-4o-mini", messages, tools: openAITools(tools) });
  const message = choices[0].message;
  messages.push(message);
  if (!message.tool_calls?.length) break;
  messages.push(...(await runOpenAIToolCalls(tools, message.tool_calls)));
}

A runnable version without the OpenAI package is in examples/adapters/openai-chat.ts.

OpenAI Agents SDK

import { Agent, run, tool } from "@openai/agents";
import { openAIAgentsTools, spliceTools } from "@spliceloom/adapters";

const agent = new Agent({
  name: "Research",
  instructions: "Answer from the tools and name the source of every number.",
  tools: openAIAgentsTools(await spliceTools(), tool),
});
console.log((await run(agent, "Top Robinhood stock tokens today?")).finalOutput);

LangChain

import { tool } from "@langchain/core/tools";
import { createReactAgent } from "@langchain/langgraph/prebuilt";
import { ChatOpenAI } from "@langchain/openai";
import { langChainTools, spliceTools } from "@spliceloom/adapters";

const agent = createReactAgent({ llm: new ChatOpenAI({ model: "gpt-4o-mini" }), tools: langChainTools(await spliceTools(), tool) });

Vercel AI SDK

import { generateText, jsonSchema, stepCountIs, tool } from "ai";
import { openai } from "@ai-sdk/openai";
import { aiSdkTools, spliceTools } from "@spliceloom/adapters";

const { text } = await generateText({
  model: openai("gpt-4o-mini"),
  tools: aiSdkTools(await spliceTools(), { tool, jsonSchema }),
  stopWhen: stepCountIs(5),
  prompt: "Which perpetual markets on Robinhood Chain have the most open interest?",
});

What stays the same

• Skills run in the Splice sandbox, limited to the permissions they declared, exactly as with splice run. Install them first with splice add <package>.
• Data tools validate their input against the schema before any provider is contacted.
• Provider keys stay in your process. They are read from the provider environment (.env.local, ~/.splice/.env) and redacted from every result; the model only sees data.
• Every data result keeps its status and provenance (LIVE, CACHED, UNAVAILABLE, ERROR, source, fetch time). Results passed back to the model are trimmed to about 9,000 characters; execute() returns the full result.