SandBase Docs

One API surface for LLMs, image, video, audio, embeddings, real-world APIs, and reusable Agents.

Docs · Store · Console · CLI / MCP · Blog

---

SandBase helps developers build AI applications and Agent workflows without maintaining a separate integration for every model, media generator, search provider, or data API.

Use the smallest surface that fits the job:

Need	Start with
OpenAI-compatible LLM and vision calls	LLM Gateway
Image, video, audio, or embedding jobs	Model API Reference
Search, scrape, social, business, or other external data	Store
Reusable, versioned Agent workflows	Agents
Codex, Claude Code, Cursor, and other AI clients	CLI setup

The catalog, schemas, prices, and availability can change. Use the live Store and API reference as the source of truth instead of guessing model or API identifiers.

Open-source companion projects

The documentation is part of a small, interoperable open-source stack:

• SandBase CLI — connect supported AI clients to the SandBase catalog through a local MCP bridge.
• SandBase Skills — install reusable research, marketing, and business workflows into compatible agents.
• DeepSeek Harness Handbook — use source-backed guides for operating and extending an agent runtime.

Choose the project that matches the layer you are building; the links above are maintained as the canonical contribution and issue entry points.

Build and contribute locally

The published site is built from this repository with VitePress. To preview a documentation change before opening a pull request:

npm ci
npm run dev

For a production-equivalent check, run npm run build; the generated site is written to .vitepress/dist. Keep model IDs, request fields, and pricing tied to the linked live reference pages, and open an issue when the source of truth is unclear. See the repository's open issues for current deployment and content work.

First LLM call

Create an API key in the Console, export it, and select a current model ID from Supported Models.

export SANDBASE_API_KEY="sk-sb-YOUR_API_KEY"

curl https://api.sandbase.ai/v1/chat/completions \
  -H "Authorization: Bearer $SANDBASE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek/deepseek-v3",
    "messages": [
      {"role": "user", "content": "Explain this API response."}
    ]
  }'

OpenAI SDK applications can use the same endpoint by changing the API key, base URL, and model:

import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ["SANDBASE_API_KEY"],
    base_url="https://api.sandbase.ai/v1",
)

response = client.chat.completions.create(
    model="deepseek/deepseek-v3",
    messages=[{"role": "user", "content": "Explain this API response."}],
)

print(response.choices[0].message.content)

Compatibility reduces migration work; it does not make every model, parameter, streaming event, or provider-specific feature identical. Follow the first-call guide and test the behaviors your application actually uses.

Run a model or external API

Image, video, audio, embedding, and external API operations use the public run contract documented on each Store detail page. The model/API identifier and input fields are operation-specific.

curl -X POST https://api.sandbase.ai/v1/run \
  -H "Authorization: Bearer $SANDBASE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "YOUR_CURRENT_MODEL_OR_API_ID",
    "YOUR_REQUIRED_INPUT": "value"
  }'

Do not force unlike workloads into one response lifecycle: an LLM may stream tokens, a synchronous data API may return immediately, and a video job may require polling or webhook handling. Check the selected capability's schema and execution mode before integrating it.

Connect an AI client

The open-source SandBase CLI provides a local MCP bridge for supported AI clients. Use the versioned installation command from the CLI setup guide, then connect the exact client target listed there.

Skills and execution remain separate:

• SandBase Skills package repeatable workflows.
• The CLI/MCP bridge exposes discovery and execution tools to a compatible client.
• SandBase Harness is the local-first runtime when Agents need persistent sessions, sandboxed tools, credentials, audit, and replay.

Documentation map

Section	What it covers
Getting Started	API keys, quickstart, and first call
Store	Models, APIs, Agents, and Skills
Models	Supported interfaces and capability guidance
API Reference	Authentication, endpoints, schemas, errors, and webhooks
Agents	Reusable Agent definitions, Services, Sessions, and Schedules
Setup	Connect AI clients and install capabilities
For AI Tools	Machine-readable integration guidance
Guides	Streaming, errors, rate limits, and billing

Repository structure

getting-started/       # first-call onboarding
store/                 # catalog concepts
models/                # model interfaces and capabilities
model-api-reference/   # generated model and external API references
api-reference/         # REST API contracts
agents/                # reusable Agent workflows and services
setup/                 # CLI and client setup
for-agents/            # concise machine-readable guidance
guides/                # operational practices

Develop locally

npm ci
npm run test:generator
npm run build
npm run dev

npm run test:generator and the model/API reference generator read the versioned catalog from the sibling sandbase-registry checkout at ../sandbase-registry/data/. The public Docs repository does not vendor that catalog. If you only need to preview already-generated pages, npm run build does not require the sibling checkout; if you are changing generated references, prepare the matching registry checkout first and run the generator before committing its outputs.

The site is built with VitePress. Production deployment is owned by this repository; see DEPLOYMENT.md.

Contributing

Found a missing capability, broken link, unclear schema, or outdated example? Open an issue or submit a focused pull request. Do not include API keys, customer data, or provider credentials in examples.

License

Apache-2.0