# SandBase Quickstart — give this to your AI agent

Use this guide with the [SandBase setup Skill](https://www.sandbase.ai/SKILL.md). It extends the [existing Quickstart](https://www.sandbase.ai/docs/getting-started/quickstart): account and workspace, choose a connection, make a real request, then build the feature you asked for.

## Start in your AI assistant's chat

Paste this sentence into the chat of your AI coding assistant, not into a terminal:

```text
set up https://www.sandbase.ai/SKILL.md
```

Then describe what you need. You can include a SandBase API page and an existing project. Do not include an API key or password.

Example to adapt:

```text
Help me build a website traffic lookup using SandBase Similarweb:
https://www.sandbase.ai/apis/similarweb

I already have a SandBase account and API key. Inspect my current project,
reuse its framework, and ask for any missing domain, period, or metric.
Find the right endpoint, check its inputs and price, and implement the feature.
Show the proposed live-test cost before making a paid request.
Keep my key in the local/server environment, never in chat or browser code.
Give me the run command and tell me what you actually tested.
```

## Choose the path that matches your situation

| Your situation | Next step |
|---|---|
| No account | The setup flow guides you to sign in with GitHub or Google. First login creates a personal workspace; creating a team is optional. |
| Already have a key | Tell the agent it exists. Reuse local credentials; do not paste the value into chat or create another key unnecessarily. |
| Already connected an AI tool | Ask the agent to check its existing tools and continue with your task. Reinstalling is not the first step. |
| Existing project | Open that project in your coding assistant and describe the feature. The agent should extend your project, not replace it. |
| No project yet | Describe the result you need: a script, an app feature, or just a dataset. The agent should clarify only what is missing. |
| Want to call HTTP directly | Follow [First API call](https://www.sandbase.ai/docs/getting-started/first-call) and the chosen API's documentation. MCP or a Skill is not required in your deployed app. |
| Want to try it in a browser | Sign in, choose an endpoint from the [catalog](https://www.sandbase.ai/apis), check its price and parameters, and run one request on its page. |
| Want a hosted Agent | Follow [Build an Agent](https://www.sandbase.ai/docs/agents/). Installing this setup Skill connects your existing assistant; it does not create a hosted Agent. |

For AI-tool setup, follow browser authorization when prompted; you do not need to paste an API key into the conversation. Confirm the workspace you want to use. Start a new assistant session if the tools need to reload. If your assistant cannot run commands or load tools, use [client-specific setup](https://www.sandbase.ai/console/setup).

For HTTP development, create or reuse a key in the intended workspace under [API Keys](https://www.sandbase.ai/console/keys). Configure `SANDBASE_API_KEY` in your local/server environment or secret store. Do not put the key in browser JavaScript or commit it to a repository.

## Instructions for the agent

Read the [setup Skill](https://www.sandbase.ai/SKILL.md) for connection commands, discovery/inspection, execution, polling, and error handling. If you already read it, continue from the user's current state instead of restarting setup. Apply its quick-start decision table and development workflow to the user's actual goal.

- Reuse their account, workspace, connection, project, and supplied API URL. Ask only for missing information.
- Use discovery and inspection to select the right tool and check its live input schema and price. A platform directory is not an endpoint. Do not assume every data API is free.
- Implement the requested integration and its relevant failure handling. For an existing app, follow its conventions and keep API credentials server-side.
- Run a small live test within the user's authorization and budget. If it cannot be run, complete offline work and clearly separate tested behavior from pending live verification. Never label mock output as a real result.
- Return the runnable feature or requested data, how to use it, actual validation, and any next step. A setup-only request does not authorize an invented app, billed demo, or deployment.

## Know when it works

Creating a key, installing a Skill, and seeing a run ID are intermediate steps. Distinguish configuration, usable tools, a verified business call, and a working feature. Show the actual output or saved file and explain empty/error responses. For code, provide the start command and verify the intended user flow.

If a call fails, check the active workspace and authentication first, then read the reported error. A key spending limit and insufficient workspace balance are different issues. Consult [Errors and retries](https://www.sandbase.ai/docs/guides/error-handling) and [Billing](https://www.sandbase.ai/console/billing) as applicable; do not blindly repeat a billed run.

The [full onboarding guide](https://www.sandbase.ai/docs/getting-started/quickstart) covers account creation, optional teams, website use, direct API calls, AI-tool setup, and hosted Agents. This Markdown guide can be shared with your agent as a URL or a file; it contains no credentials.