One API Key. Build Your Personal Agent.

2,000+ models & APIs · 400+ AI models · 1,800+ data APIs from 100+ platforms
100 free API calls / day · Free sandboxes · Free agents

Get an API key · All models & APIs · By platform · Docs · 中文

---

Your personal agent in one call

Create an agent, give it a job, get the answer. Every agent can find and call any of the 2,000+ models and APIs on SandBase on its own, and it works in its own sandbox. Agents and sandboxes are free. You pay only for the models and APIs your agent uses.

# pip install "git+https://github.com/sandbaseai/sandbase#subdirectory=sdk/python"
from sandbase import SandBase

sb = SandBase()  # reads SANDBASE_API_KEY

agent = sb.create_agent(
    "deepseek/deepseek-v4-flash",
    instructions="You are my research agent. Use SandBase tools for live data.",
)

reply = sb.run_agent(
    agent["id"],
    "What are people on X saying about AI agents today?",
)
print(reply["output"])

In our test run the agent searched for an X API, read its schema and price, called it, and answered with three real posts and their links. It took 45 seconds and needed no setup beyond the key.

How the agent gets its tools

Every new agent comes with the SandBase tools (discover, inspect, run) by default. You don't need to configure anything. If you set mcp_connections yourself, the list replaces the default:

# add your own MCP server and keep the SandBase tools
mcp_connections=[
    {"connection_id": "builtin:sandbase", "server_label": "sandbase_builtin", "required": True},
    {"connection_id": "mcp_...", "server_label": "github", "required": True},
]

# a plain chat agent without tools
mcp_connections=[]

The built-in entry must use server_label: "sandbase_builtin". Any other label returns 400.

Keep the conversation going with sb.send(reply["session_id"], "..."). When the agent is ready, deploy it as an HTTP Service (sb.invoke_service(...)) or put it on a Schedule. Agents follow the OpenAI Agents API, so the official OpenAI SDK works as well (client.beta.agents, example). TypeScript and Go have the same helpers: runAgent and RunAgent.

What your agent gets

		Price
🧰 A toolkit	2,000+ models and APIs: LLMs, image, video and audio models, plus social, search, SEO, e-commerce and finance data from 100+ platforms	Pay as you go · 100 free API calls / day
📦 A place to work	An isolated sandbox for code, commands and files, compatible with the E2B SDK	Free
🤖 A job	Sessions, HTTP Services and Schedules on the OpenAgentCore runtime, called through the OpenAI Agents API	Free

Pick your path

You are	Start here
Building a personal agent	The example above and the Python · TypeScript · Go clients
Using Claude Code, Codex or Cursor	Paste into your agent: npm install -g @sandbaseai/cli && sandbase connect (CLI / MCP)
Calling a model or API from your app	Call the pieces directly
Running a gateway or relay (new-api, one-api, LiteLLM)	Relay guide and price files
An API provider	Add your API with a pull request

Call the pieces directly

Everything your agent uses is also available to your own code with the same key.

Any model or API. Every capability takes the same request: model plus its own fields.

curl https://api.sandbase.ai/v1/run \
  -H "Authorization: Bearer $SANDBASE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "exa/search", "query": "AI agents"}'

sb.run("exa/search", query="AI agents")  # a data API (sync)

sb.run(                                  # a video model (async, run() waits)
    "bytedance/seedance/2.5/text-to-video",
    prompt="a cat surfing",
    duration=4,
)

A sandbox. Keep the official E2B SDK and point it at SandBase:

# pip install e2b
import os
from e2b import Sandbox

os.environ["E2B_API_KEY"] = os.environ["SANDBASE_API_KEY"]
os.environ["E2B_API_URL"] = "https://sandbox.sandbase.ai"
os.environ["E2B_DOMAIN"] = "sandbox.sandbase.ai"

sbx = Sandbox.create("tpl_...", timeout=120)  # template ids: catalog/sandbox.json
print(sbx.commands.run("python3 -c 'print(40 + 2)'").stdout)
sbx.kill()

TypeScript

// sdk/typescript: one file, no dependencies
import { SandBase } from "@sandbaseai/sdk";

const sb = new SandBase();
const result = await sb.run("exa/search", {
  query: "AI agents",
});

Go

import sandbase "github.com/sandbaseai/sandbase/sdk/go"

c, _ := sandbase.NewClient("") // reads SANDBASE_API_KEY
run, err := c.Run(ctx, "exa/search", sandbase.Params{
    "query": "AI agents",
})

LLMs with the OpenAI SDK

import os
from openai import OpenAI

client = OpenAI(
    base_url="https://api.sandbase.ai/v1",
    api_key=os.environ["SANDBASE_API_KEY"],
)
client.chat.completions.create(
    model="openai/gpt-6.1-sol",
    messages=[{"role": "user", "content": "Hi"}],
)

The Anthropic Messages (/v1/messages) and Gemini (/v1beta/models/...) formats are also accepted.

HTTP file (VS Code / JetBrains)

Open examples/http/sandbase.http and click Send Request. It covers sync and async runs, costs, images, video with webhooks, LLMs, Agents and sandbox templates.

What's inside

	Count	Includes
AI models	400+	100+ LLMs, 150+ video, 100+ image, 30+ audio: Alibaba, Google, OpenAI, Kling, ByteDance, MiniMax, Black Forest Labs, xAI, …
Social & content	1,200+	Douyin, TikTok, Instagram, Xiaohongshu, Weibo, YouTube, Zhihu, Bilibili, Kuaishou, LinkedIn, …
Search, SEO & web data	200+	DataForSEO, Similarweb, Brave, Google, Firecrawl, Artificial Analysis, Bing, DuckDuckGo, Exa, Scholar, …
E-commerce, apps & reviews	100+	1688, GetApp, TikTok Shop, Amazon, Clutch, Google Play, Trustpilot, G2, JD.com, Taobao, …
Finance, crypto & markets	150+	Polymarket, FRED, Financial Datasets, CoinGecko, Nasdaq, MarketBeat, SECForm4, StockAnalysis, Yahoo Finance, Kalshi, …
Companies, people & jobs	40+	Web3.career, Apollo, Levels.fyi, Y Combinator, Wellfound, Crunchbase, Hunter, Built In, Indeed, …
More	70+	Semrush, IMDb, AAA, Open-Meteo, IQAir, AccuWeather, Yahoo, Yandex, Baidu, China Social Media, …

Every entry has a page with its schema, price and code in four languages. See all models & APIs or browse by platform: TikTok · Instagram · YouTube · LinkedIn · Reddit · Douyin · Xiaohongshu · DataForSEO · Polymarket · OpenAI · Google · Kling

Pricing

Every model record in catalog/models/<name>.json has a structured pricing field. If the price depends on parameters, a precomputed price table shows the cost before you call:

// catalog/models/openai/gpt-image-2.5-flare.json → pricing.price_examples
{ "params": { "quality": "low",  "resolution": "1k" }, "usd": 0.01 }
{ "params": { "quality": "high", "resolution": "2k" }, "usd": 0.15 }
{ "params": { "quality": "max",  "resolution": "4k" }, "usd": 1 }

• Free: sandboxes, Agents, and 100 data-API calls per organization per day (resets 00:00 UTC).
• Pay as you go: everything else, at the list price in USD. Failed calls are not charged.
• Actual cost of a run: sb.cost(id) or GET /v1/tasks/{id}/cost.
• Price history: the catalog refreshes daily, so git log -p catalog/ shows every price change.

Pricing models

pricing.model	Meaning
per_call	Flat price per successful call
per_token	per_million_tokens.input / output, plus cache multipliers
formula	Depends on parameters; see price_examples and reference_price
usage_based	Metered per call; reference_price is typical
free	No charge

Sync, async and errors

• sync: POST /v1/run returns 200 with status: "completed" and the result in output / outputs.
• async: POST /v1/run returns {"id", "status": "pending"}. Poll GET /v1/run/{id}, or pass webhook_url (image, video, audio).
• Only send a mode listed in the model record. Unsupported modes return 400.
• 400 invalid input (checked against input_schema before any charge) · 401 bad key · 402 insufficient balance · 429 rate or free-quota limit.

Repository layout

Path	Contents
catalog/index.json	Every model and API: name, type, modes, endpoint, price summary
catalog/models/<name>.json	Full record: description, input_schema, output schema, modes, pricing, examples
catalog/skills.json, catalog/sandbox.json	Public Skills; sandbox connection settings and templates
docs/	One page per model and per platform, with code in four languages
openapi/sandbase.openapi.json	OpenAPI 3.1. Import it into Postman, Apifox, gateways or SDK generators
sdk/python, sdk/typescript, sdk/go	Dependency-free clients for models, APIs and Agents
examples/	.http, curl, sandboxes (E2B), Agents (OpenAI SDK)
exports/	Price and model files for new-api / one-api, LiteLLM, OpenRouter-format tools, and CSV
llms.txt	A map of this repository for AI assistants

FAQ

Is SandBase OpenAI-compatible? Yes. Chat Completions, Responses and the Agents API work with the official OpenAI SDKs at https://api.sandbase.ai/v1.

Can I keep my E2B code? Yes. Change E2B_API_KEY, E2B_API_URL and E2B_DOMAIN.

How do I get TikTok, Instagram, YouTube, Douyin or Xiaohongshu data? Find the endpoint on its platform page and call POST /v1/run with its model name. You don't need an account with each platform.

What is free? Sandboxes, Agents, and 100 data-API calls a day.

Can AI coding agents use this? Yes. SandBase CLI / MCP connects Claude Code, Codex, Cursor and 22 more. AI assistants can also read llms.txt.

Add your API

API owners can propose an integration with a pull request under integrations/. CI validates it, a maintainer reviews it, and merging starts an automated rollout to every SandBase key.

Contributing

Found a wrong field or price? Open a catalog issue. SDK and example pull requests are welcome; see CONTRIBUTING.md. If SandBase saves you an integration, a ⭐ helps others find it.

Code is licensed under Apache-2.0. For the catalog data, see NOTICE.md.