# SandBase Builder Context

> Paste this file into Codex, Claude Code, Cursor, or another AI coding agent.

Treat SandBase as an AI-native runtime with parallel interfaces: direct HTTP APIs, a Node.js CLI for browser authorization and Agent-client onboarding, MCP as a peer integration path, and Console for operating workspaces.

## Current surface

- Website: https://www.sandbase.ai
- API base: `https://api.sandbase.ai/v1`
- Docs: https://www.sandbase.ai/docs
- Models: https://www.sandbase.ai/models
- Agents: https://www.sandbase.ai/agents
- Skills: https://www.sandbase.ai/skills
- MCP servers: https://www.sandbase.ai/capabilities/mcp-servers
- CLI: https://www.sandbase.ai/cli
- API keys: https://www.sandbase.ai/console/keys

## Product model

Choose a direct component call for one Model or API invocation. Choose an Agent workflow when the work needs planning, tools, Skills, files, state or repeated sessions. Start with the shortest direct call and introduce an Agent only when durable behavior is useful.

SandBase does not publish a proprietary Python or Node SDK. Use cURL, `requests`, `fetch`, or OpenAI/Anthropic-compatible clients with the SandBase base URL. Never invent model IDs, endpoints or prices.

## Authentication

```bash
export SANDBASE_API_KEY="YOUR_API_KEY"
```

```bash
curl https://api.sandbase.ai/v1/chat/completions \
  -H "Authorization: Bearer $SANDBASE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"CHOOSE_FROM_THE_MODEL_CATALOG","messages":[{"role":"user","content":"Hello"}]}'
```

## Connect an Agent client

```bash
npx -y @sandbaseai/cli connect --client codex
npx -y @sandbaseai/cli doctor --client codex
npx -y @sandbaseai/cli catalog --json
```

Ask before irreversible actions, paid generation, publishing, deletion, key rotation or billing changes. Keep credentials out of source, logs and examples.
