Start building
SandBase

Appearance

Frequently Asked Questions ​

What is SandBase? ​

SandBase helps Agents connect to the real world.

You can use ready-made Models and APIs, use them from your AI tools after connecting them in AI tools, or build reusable Agents and run them in Sessions.

Who is SandBase for? ​

SandBase is mainly for three groups:

  • AI tool users who want useful capabilities inside Codex, Claude, Cursor, or Kiro
  • builders who want to call Models and APIs directly
  • FDEs and teams who build reusable Agents for customers or internal workflows

Where do I browse capabilities? ​

The SandBase website lists the catalog of capabilities:

What is AI tools? ​

Build → AI tools in the Console connects SandBase to the AI tool you already use, such as Claude Code, Codex, Cursor, or Kiro. Paste set up https://www.sandbase.ai/SKILL.md into your agent, or run npx -y @sandbaseai/cli@latest connect. The connected tool can search and run the SandBase catalog of Models and APIs, billed to your workspace. See Connect AI tools.

What is Build Agent? ​

Build Agent is where you define and test an Agent.

An Agent combines instructions, a Model, Skills, and MCP tools into one reusable, versioned workflow.

Is an Agent different from an API? ​

Yes. An Agent is a versioned configuration that you edit and test, and you run it by creating a Session. Your app can create Sessions with the Sessions API while you keep editing the Agent.

What are Sessions? ​

A Session is one Agent execution with its conversation, tool activity, and output.

How do I give an Agent access to external APIs? ​

Connect the external system as an MCP server and add it to the Agent's MCP connections. If the server needs a Bearer token, store it with the Secrets API as a write-only Secret bound to that server URL. Tokens are sent only to that server and never appear in prompts or the Agent runtime. See Credentials for MCP tools.

Where do Models live? ​

Models have their own catalog because they are important building blocks.

You can call a Model directly, use it from a connected AI tool, or use it inside an Agent.

Do I need to write code? ​

Not always.

Connect your AI tool if you want capabilities inside it without writing code.

Use API Reference if you want to call SandBase from your application.

Use Build Agent if you want reusable multi-step work.

How do I get an API key? ​

Open the Console, go to Developer → API Keys, and select New key.

Use the key as:

http
Authorization: Bearer sk-YOUR_KEY

How does billing work? ​

SandBase uses credits and usage-based billing.

Models, APIs, Agents, and runtime work may have different costs. Check the Console for current usage and pricing.

Next steps ​