Start building
SandBase

Appearance

SandBase Docs ​

Build with Models and APIs directly, or define reusable Agents that run on managed infrastructure.

SandBase provides two complementary ways to build AI applications. Start with the smallest surface that fits the work; you can combine them as the application grows.

Choose how to build ​

Models and APIsManaged Agents
What it isDirect access to language, image, video, audio, and third-party APIsReusable, versioned Agent configurations executed by SandBase
Best forCustom application loops and fine-grained request controlMulti-step, asynchronous, scheduled, or repeatable work
You managePrompts, orchestration, retries, and application stateAgent definition, enabled capabilities, inputs, and delivery behavior
Start hereModels and APIsBuild Agent

One SandBase API key

The same API key can call Models, APIs, Agents, Services, and scheduled workloads. Keep keys on the server and never expose them in browser code.

Core concepts ​

Models and APIs ​

Use Models for language, image, video, and audio workloads. Use APIs when the application needs external data or actions. The Store shows available capabilities, pricing, input schemas, and working examples.

Agents ​

An Agent is a reusable, versioned configuration containing a Model, instructions, Skills, and MCP connections. Define it once, test it in a Session, and then choose how it should run.

Services and Sessions ​

A Service makes a pinned, tested Agent version callable by applications and AI tools. Each invocation through /v1/services creates a Run, and each Run creates its own Session with the full conversation.

Schedules and run history ​

Schedules run a pinned Agent version on a cron expression or on demand through /v1/schedules. Every firing or manual trigger creates a Run with its own ses_* Session, so you can review each result.

First steps ​

  1. Create an organization and API key.
  2. Choose a Model, API, or Agent from the Store.
  3. Make a direct request or define and test an Agent.
  4. Deploy the tested Agent version as a Service when an application needs to invoke it.
  5. Add a Schedule when the work should run automatically.
  6. Review Sessions, errors, latency, and cost in the Console.

Operational guidance ​

Next steps ​