Appearance
Framework integrations
Usefully is OpenAI-compatible, so frameworks with an OpenAI-compatible client work out of the box. Each snippet below is tested against the API.
Vercel AI SDK
Use the @usefully/ai-sdk provider:
bash
npm install @usefully/ai-sdk aits
import { generateText, streamText, embedMany } from "ai";
import { usefully } from "@usefully/ai-sdk"; // reads USEFULLY_API_KEY
const { text } = await generateText({
model: usefully("anthropic/claude-sonnet-5-5"),
prompt: "Write a haiku about failover.",
// Optional routing preferences:
providerOptions: { usefully: { provider: { order: ["anthropic"] } } },
});
const result = streamText({ model: usefully("anthropic/claude-sonnet-5-5"), prompt: "Count to five." });
for await (const part of result.textStream) process.stdout.write(part);
const { embeddings } = await embedMany({
model: usefully.embeddingModel("voyage/voyage-3.5"),
values: ["first", "second"],
});createUsefully({ apiKey, baseURL, headers, fetch }) makes a configured instance.
LangChain (Python)
bash
pip install langchain-openaipython
import os
from langchain_openai import ChatOpenAI, OpenAIEmbeddings
llm = ChatOpenAI(
base_url="https://api.usefully.info/v1",
api_key=os.environ["USEFULLY_API_KEY"],
model="anthropic/claude-sonnet-5-5",
# Optional routing preferences:
extra_body={"provider": {"sort": "latency"}},
)
print(llm.invoke("Hello!").content)
embeddings = OpenAIEmbeddings(
base_url="https://api.usefully.info/v1",
api_key=os.environ["USEFULLY_API_KEY"],
model="voyage/voyage-3.5",
# Required for non-OpenAI models: send text, not tiktoken token IDs.
check_embedding_ctx_length=False,
)LangChain (JavaScript)
ts
import { ChatOpenAI, OpenAIEmbeddings } from "@langchain/openai";
const llm = new ChatOpenAI({
model: "anthropic/claude-sonnet-5-5",
apiKey: process.env.USEFULLY_API_KEY,
configuration: { baseURL: "https://api.usefully.info/v1" },
});LlamaIndex
bash
pip install llama-index-llms-openai-like llama-index-embeddings-openai-likepython
import os
from llama_index.llms.openai_like import OpenAILike
from llama_index.embeddings.openai_like import OpenAILikeEmbedding
llm = OpenAILike(
api_base="https://api.usefully.info/v1",
api_key=os.environ["USEFULLY_API_KEY"],
model="anthropic/claude-sonnet-5-5",
is_chat_model=True,
context_window=128_000,
)
embed_model = OpenAILikeEmbedding(
api_base="https://api.usefully.info/v1",
api_key=os.environ["USEFULLY_API_KEY"],
model_name="voyage/voyage-3.5",
)Use OpenAILike (not OpenAI): the OpenAI class rejects model names it doesn't know.
CrewAI
bash
pip install "usefully[crewai]"python
from crewai import Agent
from usefully.integrations.crewai import usefully_tools
researcher = Agent(
role="Researcher",
goal="Answer questions with current, cited sources",
backstory="You check facts on the web before answering.",
tools=usefully_tools(), # web search, page reading, deep research
)usefully_tools() reads USEFULLY_API_KEY. To use a different client, pass usefully_tools(Usefully(api_key=…)). For agents that should stay fast and cheap, pass include_research=False to drop the deep research tool.
n8n
Install the community node n8n-nodes-usefully (Settings → Community Nodes), then add a Usefully API credential. The Usefully node can:
- chat with any model
- search the web
- read pages
- run deep research, waiting for the cited report
- embed text
Each input item becomes one request.